@nntaleb NEW PROBABILITY/STATISTICS PUZZLE
A casino has 10 Roulette wheels. You have learned that one of the wheels is defective and has one of its 38 numbers occur 10% too often.
If you knew which number and which wheel, you could make a fortune. How much data do you need?

More precisely: how big must N be so that after N spins of each wheel, the most frequently occurring number/wheel pair has a 90% chance of being the defective one?

You could solve this pretty easily by simulation, but I would prefer you just try to ballpark it and get within a factor of two or so by using standard approximations.

@nntaleb Nassim, part of this is EVT which you know better than I do. If you are taking the maximum of n independent standard normal variables (mean 0 SD 1), for large n, what do the mean and SD of that look like?