Crowd Lab · Prediction scorecard

The Prediction Scorecard

Anyone can claim their crowd calendar is “AI-powered”. We grade ours in public instead. Every night we compare what we Projected and what our Model said against what was actually Observed — and publish the receipt, including the days we got wrong.

Observedgraded nightly against recorded waitsModelwait model vs Disney’s posted times

Last 30 days, all parks

Observed
Crowd level within ±1
41%
ObservedProjected 1–10 crowd level within one decile of the observed day, across 180 scored park-days.
Crowd level exact
11%
ObservedProjected decile matched the observed decile exactly.
Wait model vs posted
9.3 min
ModelMean absolute error against Disney's posted waits over 497,276 model-vs-posted checks.

Sample: 497,276 model-vs-posted checks · 180 days observed · window Oct 13, 2026 – Nov 11, 2026 · updated Nov 13, 2026.

Trailing 90 days: 41% within ±1 decile (540 park-days), wait MAE 10.2 min across 1,161,253 checks.

Evidence

Accuracy by park

Projected crowd levels graded against Observed deciles; wait error is our live Model against Disney’s posted times.

Crowd and wait accuracy by parkObserved
Prediction accuracy by park: crowd-projection hit rates and model-vs-posted wait error over the trailing 30 and 90 days.
ParkWindowWithin ±1 decileExact decileDays scoredWait MAEModel checks
Disney California Adventure30 days47%10%3010 min86,196
90 days40%14%9011.7 min201,335
Disneyland30 days33%7%308.4 min147,913
90 days31%10%909.7 min347,147
Animal Kingdom30 days37%30%3010.8 min32,429
90 days32%17%9011.1 min74,374
EPCOT30 days57%0%3011.8 min60,675
90 days56%16%9012.3 min140,061
Hollywood Studios30 days47%10%3013.2 min39,371
90 days44%13%9014.2 min92,770
Magic Kingdom30 days27%7%307.3 min130,692
90 days44%26%907.5 min305,566
Evidence

The day-by-day receipt

Each cell is one park-day: projected → observed crowd decile. Green = exact, amber = ±1, red = further off. “·” means that side doesn’t exist for the day (no projection possible, or the day wasn’t observed) — those days are excluded from the rates, never counted as hits.

Disney California Adventure

View this receipt as a table
Day-by-day prediction receipt for Disney California Adventure: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 292/10 (Very light)1/10 (Very light)19.8 min3,159
Oct 305/10 (Moderate)7/10 (Busy)26.1 min2,830
Oct 317/10 (Busy)1/10 (Very light)613.6 min3,016
Nov 11/10 (Very light)1/10 (Very light)012.5 min2,585
Nov 27/10 (Busy)6/10 (Moderate)16.9 min2,810
Nov 32/10 (Very light)5/10 (Moderate)310.1 min2,718
Nov 45/10 (Moderate)5/10 (Moderate)07.4 min2,612
Nov 55/10 (Moderate)6/10 (Moderate)15.6 min2,682
Nov 65/10 (Moderate)2/10 (Very light)39.5 min2,751
Nov 75/10 (Moderate)2/10 (Very light)311.2 min2,862
Nov 83/10 (Light)1/10 (Very light)210.7 min2,846
Nov 95/10 (Moderate)5/10 (Moderate)07.7 min2,811
Nov 105/10 (Moderate)9/10 (Packed)47.8 min2,726
Nov 115/10 (Moderate)10/10 (Packed)58.7 min2,639

Disneyland

View this receipt as a table
Day-by-day prediction receipt for Disneyland: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 297/10 (Busy)3/10 (Light)410 min4,877
Oct 306/10 (Moderate)5/10 (Moderate)17.4 min5,183
Oct 317/10 (Busy)4/10 (Light)38.9 min4,731
Nov 13/10 (Light)1/10 (Very light)213.9 min4,620
Nov 27/10 (Busy)9/10 (Packed)26.1 min4,706
Nov 37/10 (Busy)5/10 (Moderate)28.3 min4,700
Nov 47/10 (Busy)4/10 (Light)38.6 min4,884
Nov 57/10 (Busy)7/10 (Busy)06.1 min4,737
Nov 63/10 (Light)2/10 (Very light)19 min4,849
Nov 73/10 (Light)2/10 (Very light)110.3 min4,939
Nov 83/10 (Light)1/10 (Very light)212.6 min4,779
Nov 97/10 (Busy)10/10 (Packed)36.6 min5,055
Nov 107/10 (Busy)9/10 (Packed)26.6 min4,852
Nov 117/10 (Busy)10/10 (Packed)39.2 min5,039

Animal Kingdom

View this receipt as a table
Day-by-day prediction receipt for Animal Kingdom: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 296/10 (Moderate)6/10 (Moderate)010 min1,092
Oct 306/10 (Moderate)8/10 (Busy)28.8 min880
Oct 316/10 (Moderate)6/10 (Moderate)014.6 min1,106
Nov 16/10 (Moderate)6/10 (Moderate)011.2 min1,056
Nov 26/10 (Moderate)8/10 (Busy)29 min1,071
Nov 36/10 (Moderate)8/10 (Busy)26.9 min1,050
Nov 46/10 (Moderate)8/10 (Busy)210.4 min1,139
Nov 56/10 (Moderate)8/10 (Busy)27.8 min1,107
Nov 66/10 (Moderate)8/10 (Busy)29.3 min1,073
Nov 76/10 (Moderate)9/10 (Packed)38.8 min1,154
Nov 86/10 (Moderate)9/10 (Packed)315.2 min1,034
Nov 96/10 (Moderate)8/10 (Busy)211.2 min973
Nov 106/10 (Moderate)6/10 (Moderate)012.5 min1,055
Nov 116/10 (Moderate)9/10 (Packed)38.9 min1,139

EPCOT

View this receipt as a table
Day-by-day prediction receipt for EPCOT: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 298/10 (Busy)9/10 (Packed)111.2 min1,847
Oct 308/10 (Busy)9/10 (Packed)111.9 min1,726
Oct 318/10 (Busy)9/10 (Packed)115.4 min2,000
Nov 18/10 (Busy)9/10 (Packed)111.3 min1,992
Nov 28/10 (Busy)9/10 (Packed)19.1 min2,068
Nov 38/10 (Busy)10/10 (Packed)29.6 min2,058
Nov 48/10 (Busy)9/10 (Packed)111.9 min2,084
Nov 58/10 (Busy)10/10 (Packed)211.1 min2,029
Nov 68/10 (Busy)10/10 (Packed)211.1 min1,834
Nov 78/10 (Busy)10/10 (Packed)210.4 min2,048
Nov 88/10 (Busy)9/10 (Packed)112.3 min2,043
Nov 98/10 (Busy)9/10 (Packed)115.3 min2,412
Nov 108/10 (Busy)10/10 (Packed)29.2 min1,880
Nov 118/10 (Busy)10/10 (Packed)29.3 min2,060

Hollywood Studios

View this receipt as a table
Day-by-day prediction receipt for Hollywood Studios: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 298/10 (Busy)8/10 (Busy)016.4 min1,301
Oct 308/10 (Busy)9/10 (Packed)111.2 min1,319
Oct 318/10 (Busy)9/10 (Packed)113.4 min1,328
Nov 17/10 (Busy)8/10 (Busy)117.2 min1,292
Nov 27/10 (Busy)9/10 (Packed)211.2 min1,286
Nov 37/10 (Busy)9/10 (Packed)212.4 min1,288
Nov 47/10 (Busy)9/10 (Packed)211.4 min1,291
Nov 57/10 (Busy)9/10 (Packed)212.3 min1,237
Nov 68/10 (Busy)10/10 (Packed)214.1 min1,284
Nov 78/10 (Busy)9/10 (Packed)111.4 min1,535
Nov 87/10 (Busy)10/10 (Packed)313.3 min1,363
Nov 97/10 (Busy)10/10 (Packed)314.1 min1,331
Nov 107/10 (Busy)9/10 (Packed)214.5 min1,299
Nov 117/10 (Busy)10/10 (Packed)311 min1,261

Magic Kingdom

View this receipt as a table
Day-by-day prediction receipt for Magic Kingdom: projected versus observed crowd decile and model-vs-posted wait error.
DateProjectedObservedDeciles offWait MAEModel checks
Oct 295/10 (Moderate)1/10 (Very light)411.3 min4,392
Oct 305/10 (Moderate)4/10 (Light)15 min4,236
Oct 319/10 (Packed)1/10 (Very light)89.7 min4,595
Nov 15/10 (Moderate)2/10 (Very light)38 min4,210
Nov 25/10 (Moderate)3/10 (Light)27.5 min3,917
Nov 35/10 (Moderate)7/10 (Busy)25.6 min4,057
Nov 45/10 (Moderate)7/10 (Busy)26.3 min3,981
Nov 55/10 (Moderate)7/10 (Busy)25.9 min4,103
Nov 65/10 (Moderate)7/10 (Busy)25.7 min4,079
Nov 75/10 (Moderate)8/10 (Busy)36.3 min4,399
Nov 85/10 (Moderate)2/10 (Very light)36.5 min4,540
Nov 95/10 (Moderate)2/10 (Very light)38.9 min4,574
Nov 105/10 (Moderate)10/10 (Packed)56.4 min4,296
Nov 115/10 (Moderate)4/10 (Light)15.8 min4,350
Candor

Our biggest misses

Biggest misses are shown, not buried: every crowd projection 2+ deciles off and the worst attraction-day model errors (at least 12 model-vs-posted checks) from the trailing 90 days, worst first.

  • Crowd projectionWe projected Packed (9/10) for Magic Kingdom on Oct 31, 2026; the day ran Very light (1/10).
  • Wait modelOur wait model averaged 83.7 min off Disney's posted wait at 412862995 (Animal Kingdom, Oct 11, 2026) — model 105 min vs posted 21 min across 68 checks.
  • Crowd projectionWe projected Packed (9/10) for Disneyland on Oct 22, 2026; the day ran Very light (2/10).
  • Wait modelOur wait model averaged 70.7 min off Disney's posted wait at Haunted Mansion Holiday (Disneyland, Aug 24, 2026) — model 103 min vs posted 32 min across 98 checks.
  • Crowd projectionWe projected Light (3/10) for Disneyland on Sep 21, 2026; the day ran Packed (10/10).
  • Wait modelOur wait model averaged 65.7 min off Disney's posted wait at Haunted Mansion Holiday (Disneyland, Aug 31, 2026) — model 87 min vs posted 22 min across 155 checks.
  • Crowd projectionWe projected Light (3/10) for Disneyland on Sep 6, 2026; the day ran Packed (10/10).
  • Wait modelOur wait model averaged 62.9 min off Disney's posted wait at Guardians of the Galaxy - Mission: BREAKOUT! (Disney California Adventure, Oct 12, 2026) — model 51 min vs posted 113 min across 72 checks.
  • Crowd projectionWe projected Busy (7/10) for Disney California Adventure on Oct 31, 2026; the day ran Very light (1/10).
  • Wait modelOur wait model averaged 62.6 min off Disney's posted wait at 412862995 (Animal Kingdom, Oct 13, 2026) — model 90 min vs posted 27 min across 72 checks.

We grade the models. The park planner puts them to work.

The park planner, free to use without signing in, runs the same crowd projections and wait models over your own date, must-rides and meals — and reoptimizes the day as the park moves.

Method

How the grading works

Crowd levels. Crowd projections are graded eve-of-day: for each park-day we replay the decile (1–10) the calendar's seasonal projection implied using only history available before that date, and compare it with the decile the observed park-wide median wait landed at on that same scale. Days without observed stats are excluded from the rates — never counted as hits. Scoring revision 2026-10-tie-aware (since 2026-10-06): each day is ranked by its park-wide median standby wait, then its average wait, and tied days share one mid-rank level.

Wait times. Wait error is the live model-vs-posted comparison: every few minutes our standby-wait model predicts each attraction's current wait and is graded against the wait Disney was posting at that same moment. MAE is the pair-weighted mean absolute error in minutes; bias is the signed mean (positive = model runs high).

The same honesty labels apply here as everywhere in the Crowd Lab: Observed means recorded from Disney’s posted waits, Typical means a historical range, Model means a machine-learning estimate shown next to observed data, and Projected means a seasonal projection for a calendar date.

Limits

What we don’t publish (yet)

Not verified

We collect wait times every minute, but we do not publish minute-level predictions for future dates. Our research nowcasts currently beat a persistence baseline at 30-, 60-, and 120-minute horizons, while the 15-minute horizon still loses to simply assuming the current wait holds — so none of them ship as public forecasts. Graduation criteria: a nowcast becomes public only when its 15-minute horizon beats persistence and it passes forward validation. Until then, future dates only ever show seasonal projections and typical ranges, labeled as such.

Keep planning

Use the predictions we do publish

  • The crowd calendar scores every park and day 1–10 — the projections this page grades.
  • The live wait-times boards show the model’s current predictions next to Disney’s posted waits, so you can judge it in real time.
  • The park planner (free to use without signing in) turns the same models into an optimized day plan.
Questions

FAQ

What is the prediction scorecard?

It is a continuously updated receipt for our own predictions. Every night we grade the previous day: the crowd level (1–10) our calendar projected for each park versus the level the observed waits actually produced, and how far our live wait-time model was from the waits Disney posted. Hit rates, error in minutes, and the biggest misses are all published here.

How is the crowd projection graded?

Eve-of-day replay: for each park-day we recompute what our seasonal projection implied using only history available before that date, then compare it with the decile the observed park-wide median wait landed at on that same scale. Days without observed data are excluded from the rates — they are never counted as hits. Scoring revision 2026-10-tie-aware (October 2026): each day is ranked by its park-wide median standby wait, then its average wait, against every day that park has on record, and tied days share one mid-rank level — so a repeated wait value never pins a whole stretch of days to a 1.

How is the wait-time model graded?

Every few minutes our standby-wait model predicts each attraction’s current wait, and each prediction is stored next to the wait Disney was posting at that same moment. The scorecard reports the mean absolute error in minutes, weighted by the number of model-vs-posted checks.

Why don’t you publish minute-level forecasts for future dates?

Because they haven’t earned it yet. Our research nowcasts beat a persistence baseline at 30–120-minute horizons, but the 15-minute horizon still loses to assuming the current wait holds. A nowcast graduates to this site only when its 15-minute horizon beats persistence and passes forward validation. Until then, future dates only show labeled seasonal projections and typical ranges.

Keep planning

Related tools

All grades are against Mickey Alerts’ own recorded observations of Disney’s posted wait times. Not affiliated with The Walt Disney Company, Disney Parks, or Disney Vacation Club.

Ready to Never Miss Out Again?

Join thousands of families who have secured their dream Disney vacations with Mickey Alerts.