There are two necessary assumptions in the problem that one might overlook when programming a simulator. It is critical that (1) the host knows where the prize is and (2) never opens that door. The original language didn't make the second assumption absolutely clear. So the simulator might make everything random and simply ignore cases in which the door with the prize is opened by the host, giving the player the choice of two wrong doors. If the host knows where the prize is, but still opens one door at random, and it happens to reveal a goat, the remaining doors each have a 50% chance of hiding the car. But I agree that hubris among PhDs was a contributing factor. I got suckered into the wrong answer too, and the standard tactic of going to extremes with a million doors helped me see the light.