Ok, maybe the croupier is not a "hidden" variable, but it could be whether he had coffee that day, which roulette wheel is being used, the lighting conditions in the room, whether he was able to park in front of the building or had to walk from the parking lot, etc. Any event that happens some of the time and causes 1, 2, 3 as the first three numbers can very well be causing a specific distribution of the next numbers.

I.e. correlation doesn't imply causation but translates to dependent probability distributions and therefore a "if A then B is likely" just because A can be observed before B.