Most wait-time apps ask you to trust them. We publish the scoreboard instead — recomputed from our own archive on a timer, bad days included.
How often the prediction for a ride landed within 5, 10 and 15 minutes of the posted wait, by park-local hour. Hours with under 30 scored predictions are withheld rather than shown from noise.
| Hour | ≤5 min | ≤10 min | ≤15 min | Median error | n |
|---|---|---|---|---|---|
| 8am | 73% | 86% | 94% | 5 min | 49 |
| 9am | 64% | 78% | 86% | 5 min | 138 |
| 10am | 59% | 76% | 85% | 5 min | 250 |
| 11am | 66% | 77% | 84% | 5 min | 309 |
| 12pm | 66% | 81% | 89% | 5 min | 271 |
| 1pm | 81% | 90% | 94% | 5 min | 93 |
| 2pm | 80% | 89% | 96% | 5 min | 93 |
| 3pm | 70% | 87% | 94% | 5 min | 94 |
| 4pm | 72% | 86% | 93% | 5 min | 92 |
| 5pm | 79% | 86% | 89% | 5 min | 209 |
| 6pm | 78% | 90% | 94% | 5 min | 204 |
| 7pm | 82% | 87% | 92% | 5 min | 218 |
| 8pm | 82% | 84% | 89% | 5 min | 418 |
| 9pm | 84% | 89% | 90% | 5 min | 401 |
| 10pm | 88% | 92% | 93% | 5 min | 318 |
| 11pm | 96% | 98% | 98% | 5 min | 231 |
| Park | ≤10 min | Median error | n |
|---|---|---|---|
| Tokyo Disneyland | 91% | 5 min | 316 |
| Hong Kong Disneyland | 84% | 5 min | 301 |
| Tokyo DisneySea | 77% | 5 min | 295 |
| Disneyland | 72% | 5 min | 247 |
| Michigan's Adventure | 100% | 5 min | 188 |
| Valleyfair | 100% | 5 min | 185 |
| Worlds of Fun | 100% | 5 min | 180 |
| Canada's Wonderland | 73% | 5 min | 155 |
| Efteling | 93% | 5 min | 140 |
| Disneyland Park | 85% | 5 min | 136 |
| California Adventure | 89% | 5 min | 123 |
| Phantasialand | 78% | 0 min | 111 |
| Six Flags St. Louis | 100% | 5 min | 85 |
| Magic Kingdom | 94% | 0 min | 84 |
| Walt Disney Studios | 52% | 10 min | 82 |
Window: 2026-08-26 to 2026-08-26 · recomputed every few hours · generated 2026-08-26
Walk-forward, no exceptions. Each day is scored using only baselines and day-of-week factors built from the days before it — the numbers the model would genuinely have shown you. A prediction never sees its own day.
One prediction per ride per hour per day. The hour's "actual" is the median of that hour's recorded snapshots. We do not score every polling tick, because that would let the sampling rate inflate the sample size.
Scored against posted waits. The model predicts the wait the park will post, so that is what it is scored against. Visitor-reported actual waits are a separate dataset with its own page per ride.
Only predictions the model was ready to make. A ride is scored once it has enough prior snapshots to carry a baseline — the same bar the product itself uses before showing a typical wait.
Everything counts. The page recomputes from the complete archive on a timer. There is no mechanism for leaving a bad day out.