This was such an interesting question. What’s the obvious way to figure out whether Bocelli sold out the Garden? You wait to see if the event organisers claim he has sold out. But what if they don’t say anything? You want another resolution method in that case, say, trying to buy tickets and seeing if there are any officially available.

Surely those two criteria will never contradict each other.

Of course they will! There were plenty of tickets remaining, and yet the organiser claimed the show was sold out. Bocelli even performed a bonus event elsewhere due to selling out the Garden. But the Garden was never sold out!

It turns out “selling out” an event is a social construction. If there are enough people there, and the relevant parties (organisers, attendants, media) reach a consensus opinion of an event having sold out, then it has sold out, regardless of ticket availability – you know, the actual technical definition of selling out.

Anyway, onto forecasting. Large parts of the community entered 99 % forecasts based on mathematical models of ticket sales in the days and weeks leading up to the show. They forgot to account for modeling error. To enter 99 % as one’s forecast based on a mathematical model means to assume that there is less than a 1 % risk that the model chosen was the wrong one for the problem. The actual risk is usually far greater than that.

In this case in particular, additional seats opened up as the available tickets ran low, violating an assumption of the mathematical model. I was fortunate enough to think of modeling error as a risk, entered a less confident forecast, and on this question it paid off.