A bunch of monkeys, each born on some day of the year. How big does the bunch need to be before it’s more likely than not that two share a birthday? Nearly everyone guesses ‘a hundred and eighty-something’. It’s 23. Slide the bunch size and watch the odds cross 50% absurdly early.
The clean way is to count the NO-match probability: the 2nd monkey must dodge 1 birthday, the 3rd dodge 2, and so on, so P(no match)=∏(1−i/365) for i=0..k−1, and P(match)=1 minus that. It climbs fast because there are k(k−1)/2 PAIRS, not k, and pairs grow quadratically. A fail-loud self-check throws unless P(1)=0, the sequence is strictly increasing, P(22)<0.5≤P(23) (the famous crossover), and P(57)>0.99.
Assumes 365 equally-likely, independent birthdays (no leap day, no twins, no seasonal clustering); real birthdays cluster slightly, which only makes matches MORE likely, so 23 is a mild upper bound. The monkeys are a wrapper; the mechanism is the birthday paradox / pair-counting.