@jdlevinson Let us get through testing the latest models this week, and I’ll do an update to the wait times next week.
I mentioned upthread that I’m headed to Ireland tonight to see the Steelers play in Dublin on Sunday. I don’t want to release new data and then leave the country. Does not endear one to one’s co-workers, you know?
We’ve tried. It’s really hard to know when a ride is going to break down, so we don’t predict it. We do detect and recover from it during your visit. There’s an “intraday” adjustment process that looks at the current day’s wait times and ride statuses, and adjust the wait-time predictions for the rest of the day. @linsalt and I talked about it on a call last week.
Part of that also tracks how long downtime has lasted at each ride, for various times of the day, and makes wait-time adjustments based on that. So, for example, we know that if Test Track breaks down within ~2 hours of park close, the odds are good it’ll stay closed for the rest of the night.
Related to this: We share data with various colleges and universities every year for student and faculty projects. We’ve been working with Wake Forest students for a couple of years now.
Last year their research question was “Knowing what we know about ride wait times and ride downtime at park opening, which two rides should we rope-drop at Hollywood Studios?”
For example, we knew that Rise of the Resistance was not running at rope drop for around 40% of the days in a given year. (I am 100% certain Disney’s internal numbers match ours.) Sending people to Rise first would result in failure for 3 out of every 7 days. That’s not great.
And that’s why we recommend people get LLSP for Rise - it’s based solely on downtime data.
Anyway, without Rise, there were 8 attractions left that you could rope-drop in DHS. That means there’s 8 x 7 = 56 possible combinations of two attractions to rope-drop.
The Wake Forest students looked at all 56 possible combinations of attractions to determine which would be the best choices, based on time saved in line and potential for downtime.
Here’s their report, if you’re interested. I thought they did a great job with the data.
Nor customers 
So at a trip not too long ago, my son and I went into HS to rope-drop RotR. He had never been on it. I did know that it could be down. But there was a CM at the entrance of the park announcing that ROTR “is currently open”. We rushed over there only to find it wasn’t running. We waited a few minutes, thinking it was a temporary thing before the CM working at the entrance to the ride told us they didn’t know when it would be up.
Fine. So, we had our back-up. Rush over to RnRC, since it was also announced by the CM as we entered the park that it was open. When we got there, it was…not open. Well, sort of. It had JUST opened, and the line was already posting I think it was a 70 minute wait. Sigh.
Turns out…the CM at the front of the park was saying that ROTR and RnRC “isn’t currently open.” Amid the noise from the sea of people during early entry, we couldn’t distinguish between her saying “is” versus “isn’t”. So, we wasted the ENTIRE early entry time going between the two rides we THOUGHT were announced as running, but actually were announced as NOT running.
In hindsight, it was rather stupid of us to think they would be publicly declaring what rides WERE open…but, as they say, hindsight has four eyes.
Oh wow 
This sounds like my life.
I’m going next Friday - hoping to see an update before we go! 
P.S. Go Steelers
Welcome to the forums! Just keep in mind the park is still new. Make sure you build a little flexibility into your plan.
Back in the office and looking at last week’s test results. Will post an update shortly.
Okay, here are the results from last week’s testing at Epic on Thursday, 9/25:
Here’s the downtime report (ignore the predicted and how-close - those numbers are from production and we’re using the models on our test system for this plan):
The park was open for 13 hours. Each hour represents 7.7% of the operating day, so 2 hours is just over 15% of the day.
Every major attraction except Ministry, Mario Kart, and Monsters had at least ~85 minutes of downtime. Some of that was weather, some of it was operational. And see below for Ministry.
Here are the attractions where the forecasts worked reasonably well:
We didn’t do great on two attractions:
Battle at the Ministry reported no downtime for the day. We got in line at 10:29 am. We’d predicted a 138-minute wait on a posted of 170 at that time.
The posted when we arrived was 150, which was reasonably close (~12%) to predicted.
The actual wait was 215, or 143% of the posted. I’m not sure how we’d model that - like, I don’t know of a way where we could reliably say “Yes, Epic’s going to say it’s 150 but it’s going to be 215” and have the general public believe us. That’s too far off. I don’t think I’d believe the model anyway. So I suspect there was some unreported downtime here.
Weather delays around 4 pm meant that’s all the testing we could do that day.
I’m updating the Hiccup and Toothless models now and am checking when we can re-test things this week.
ETA: We’ll re-test on Thursday.
Steelers actually deserved that win haha.
To be clear, I am a Steelers fan. Hope you enjoyed Ireland!
In the stands, the call to go for it on 4th and 4 from the 4 was controversial.
The running game was going well, and we were averaging 6 yards per play. Granted, 80 of those were on one play, but even without that we were still averaging a bit more than 4 yards per play. It would’ve put the game away. So I kept saying Tomlin’s “Don’t live in your fears” line. I still think it was the right decision.
I love looking at the updates and mathy bits. I won’t be able to go until December 2026 (I already have my hotels booked) but I’m really looking forward to it.
I’ve re-run the Toothless meet & greet models a number of different ways today.
The posted 85/actual 61 at 3:15 pm we observed is higher than any of the models would predict, given the data we have to this point. So I suspect the wait time here was influenced by downtime as well.
I did make some tweaks to the model to get the predicted actual wait closer to what we saw, along with the updates to Hiccup’s Wing Gliders mentioned earlier.
I’m going to load new predictions to staging tonight. Assuming they look reasonable, I’ll load them to production tomorrow.
Latest one-day plan for testing, without shows or Stardust. Omitting the shows allows us to test more ride wait-time predictions during the busy part of the day.
ETA: These new wait-time models are being loaded into production now, and should be available by 8 am on Wednesday.
Epic’s new wait-time models are live.
One thing to note: Bring along a paper copy of the plan if you’re going to Epic in the next couple of weeks.
The way we’ve been updating wait-time estimates during the day (our “intraday adjustment” process) is susceptible to large, fast swings in posted wait times, like we see at Epic when rides go offline. So you could end up with artificially high (or low) actual wait time estimates in the app.
Fixing this is my main project right now, and I don’t think it’s super-difficult. My task right now is to model and graph the existing process in a way where I can simulate the same day over and over again, in 15-minute increments, to see exactly where it goes wrong.
And we’re on for testing in the park today! I’m recording a show with Jim around 11 am and will post Epic updates as I get them.
Amy was about 30 minutes ahead of schedule this morning after completing:
Amy was headed into into Ministry of Magic just before noon with a posted wait of 135 and we’re expecting an actual of 114.
She’ll have a short break after Ministry before getting in line for Mine-Cart Madness, with an expected wait of 108 minutes.
Oh, and I used the “put free time at end of the plan” option for this day, so Amy could leave at a reasonable hour. If you’re there all day, the optimizer would likely employ a “wait out the crowds” strategy for some of these attractions.
Amy ended up waiting 254 minutes on a posted of 135 at Ministry. The posted wait remained 135 for the first hour she was in line. They didn’t report any downtime.
I’m headed out for a sec here but I’m really struggling with Epic’s posted being off by almost 100% on a 2 hour wait. Will think about this.
I love that option. Gives you a chance to do the waiting you are going to do anyway while in some line.
Amy’s last ride of the day was Mine-Cart Madness. Posted was 140. We predicted an actual of 108 and it was 82. Amy got there 2 hours later in the afternoon because of the wait at Ministry, so I suspect crowds had already started to thin a bit.
I’m going to adjust Ministry today to reflect inaccurate wait times in these situations. I think that’s the only model I’m going to tweak for now. We’ll test again next week.
