Forecast accuracy

Every morning before opening we freeze each ride's wait forecast, quarter-hour by quarter-hour, without looking at that day's data. In the evening we compare it with the actual posted waits. Here are the results, day by day.

Mean error

average gap, in minutes, between the morning forecast and the actual wait (per ride and per quarter-hour).

Naive baseline

the same measure for a simplistic forecast (typical day shape × the ride's usual level): our model has to beat it.

Within ±10 min

share of quarter-hours where the forecast is less than 10 minutes off.

Live (+1 h)

on the day we re-anchor the forecast on the current wait: error measured one hour later.

Actual / forecast

above 1 means bigger crowds than expected. Measured per type of day (normal, school holidays, bank holiday), it automatically corrects the next forecasts.

The first results come in after the first full measured day: check back tomorrow.