Park Queue Times · the crowd-forecasting model
Almanac
Almanac forecasts how busy a theme park will be on any day, up to six weeks ahead. Across more than 80 parks it is 87.3% accurate on days it was never shown, ahead of both simple ways to guess a crowd. It reads millions of live wait times, then grades itself in the open against what actually happened.
01 · The scorecard
How accurate is
the crowd forecast?
Every figure below is forecast accuracy six weeks out, measured only on real days the model was never shown. We hold Almanac up against the two simplest ways to guess a crowd. It is ahead at every park.
These are edition 1.2's figures. A new engine started forecasting on 28 August 2026, and we will not print an accuracy for it until it has been scored against enough real days to mean something — which takes a few weeks from that date. Past days on the calendar are measurements, not forecasts, and are unaffected.
| Park | Almanac | Last-week guess | Seasonal average |
|---|---|---|---|
| Thorpe ParkUnited Kingdom | 85.5% | 72.0% | 91.1% |
| Alton TowersUnited Kingdom | 86.5% | 83.3% | 83.2% |
| Legoland WindsorUnited Kingdom | 70.9% | 56.3% | 56.9% |
| Paultons ParkUnited Kingdom | 88.9% | 86.0% | 66.1% |
| Blackpool Pleasure BeachUnited Kingdom | 79.1% | 57.9% | 92.5% |
| ChessingtonUnited Kingdom | 87.2% | 88.1% | 71.6% |
| Every park, averagedall 80+ parks | 87.3% | 85.4% | 84.3% |
Accuracy = 100 minus the average miss on a 0 to 100 busyness scale, over a six-week forecast. Higher is better.
02 · Revision history
Every version of
the Almanac model
Almanac ships in versions, and each one earns its place the same way: it only goes live if a fair test proves it beats the version before.
- 82.9%The first edition v0.9.0 · February 2026
- Our starting point. It learned the broad shape of busy versus quiet days, but on a fair test it barely edged out simply repeating last week.
- 85.8%Smarter blending v1.0.0 · April 2026
- The leap that earned the 1.0 name. Almanac learned to tell a calm midweek apart from a packed half-term, and to lean on the most reliable signal for each individual park, pulling clearly ahead of the simple methods.
- 86.4%School calendar and sharper tuning v1.0.2 · June 2026
- Added UK school-term and holiday dates, so a quiet term-time weekday is no longer mistaken for a busy holiday crowd, and tuned Almanac to lean harder on each park’s most reliable signal. The most accurate Almanac yet.
- measuringA new engine v2.0.0 · August 2026 Current
- A rebuilt engine, and a stricter rule: it may only use what was genuinely knowable on the day it made the call, so its record cannot flatter itself. It began forecasting on 28 August 2026, and its measured accuracy will be published here once enough of those forecasts have been scored against real days - not before.
- 87.3%A UK edition and a rebuilt crowd scale v1.2.0 · August 2026
- UK parks now get their own model, trained only on UK data, and every past day is scored on a rebuilt scale where 100 means the park’s own busiest day. Retrained on the full summer season and re-measured against the same fair test as every edition before it.
03 · The inputs
What the crowd
forecast reads
A good forecast is more than what last year did. Almanac weighs everything that actually moves the queues.
Millions of live readings
Almanac learns from real wait times collected across more than 80 parks, updated continuously, not a one-off snapshot from last season.
The calendar
Weekends, school terms, public holidays and the time of year all change how busy a park gets. Almanac knows the difference between a sleepy Tuesday and the first day of half-term.
The weather
A washout keeps crowds at home; a warm, dry spell brings them out. Almanac nudges its near-term forecasts up or down with the forecast.
Its own track record
When a day ends, Almanac checks its forecast against what actually happened. Each park’s real history keeps tuning the next prediction.
Notes on method
It is easy for a model to look clever on data it has already seen, so we never test Almanac that way. Every version is scored only on future days it was hidden from, the same test a real visitor's trip would be.
Once a day is over, Almanac compares its forecast to what really happened across more than 80 parks. The gaps it still gets wrong become the next thing we fix, which is how each edition moves the number.
Every crowd calendar
on the site runs on this.
Pick a park and see the forecast Almanac has already made for every day of the year.