The ballroom hums. Guests arrive whenever they please — a true random stream. The organizer must find a dance partner for everyone. You can pair two up right away — but what if someone far more suitable walks in a minute later? And if you hesitate, impatience builds: standing without a partner is boring and disheartening. Such a puzzle is a headache for programs managing deliveries or cloud computing, where decisions must be made without knowing the future.

The key to success lies in how we measure the discontent of those waiting. If the penalty grows with the number of solitary guests (each extra one adds trouble), the number of possible pair combinations explodes: it grows as 2 to the power n, where n is the number of guests.

The idea of this contest against a perfect opponent was introduced by John von Neumann for analyzing strategic games. For a scale-dependent penalty, any fixed plan that doesn't adapt to the situation loses to the ideal exponentially.

But flip the impatience mode, and the problem suddenly becomes simple. Imagine that each guest's discontent grows with acceleration: the first minute of waiting is bearable, but the fifth one drives you up the wall. Add to that a tiny but mandatory penalty just for the delay itself, and the organizer's behavior changes abruptly, as if the system switched modes. The discontent formula is described as f(t) = $t^k$/k + t, where t is time. There's no penalty at zero, but at the very first moment the term t appears, and then growth accelerates. A simple rule is enough: if someone waits longer than a threshold of 4δ, immediately pair them up with the nearest partner. This 'impatient' algorithm looks only at the clock, and still works almost perfectly, losing to the all-knowing opponent by no more than a constant factor.

So the ball ends without drama: a little patience, but not too much. This simple lesson is now baked into logistics and cloud services, so that pairs, whether they be orders or tasks, meet on time.