UPDATE: Zwift greatly improved their banding algorithm after this post was published. To see current “keep together” test results, read the updated post.
Have you ever done a “Keep Together” ride in Zwift? In this mode, all riders in the event are kept in the same group, regardless of the power you’re putting out.
“Rubberbanding” makes it easy to keep a group of cyclists together, and that’s a good thing, because any ride leader will tell you that keeping cyclists together without some artificial help can feel like herding cats!
Testing the Rubberbands
So how does Zwift’s “Keep Everyone Together” logic work? I’m not entirely sure, and Zwift isn’t revealing the secret sauce. So I’ve been running tests to find out what speeds rubberbanded riders reach at different power levels, in different event setups. For science!
The core question I want to answer is: Do rubberbanded speeds make sense? Are they too slow, too fast, or some strange mix of both?
My first tests were on flat ground in a banded Club Event. Then I did more flat tests, but this time using a banded Meetup. Today, we’re digging into our first banded climb tests, using a banded Club Event.
The riders in these tests were 75kg in weight, 183cm tall, and rode Zwift Carbon bikes with 32mm carbon wheels. I timed the riders on the Alpe du Zwift climb segment since it’s steep and long enough to be a good representation of climb performance.
Test 1: A Pair of Bots
This first round of tests used just two riders in a Club event with “Keep Everyone Together” mode enabled. Here are each rider’s power outputs, segment times, and average speeds.
| Rider 1 | Rider 2 | Time | Speed (kph) |
| 300W | 300W | 47:18 | 15.5 |
| 300W | 295W | 47:52 | 15.3 |
| 300W | 225W | 40:10 | 18.3 |
| 300W | 150W | 40:10 | 18.3 |
| 300W | 75W | 40:11 | 18.2 |
| 300W | 10W | 42:46 | 17.1 |
| 225W | 225W | 1:01:13 | 12.0 |
| 225W | 150W | 50:56 | 14.4 |
| 225W | 75W | 50:56 | 14.4 |
| 150W | 150W | 1:28:46 | 8.3 |
| 150W | 75W | 1:12:59 | 10 |
Observations
The first thing that stands out here is how illogical the results are. Why would two banded riders holding 300W be 7 minutes slower than one rider holding 300W banded to a rider holding 75W?
This pattern repeats itself, too. When two riders match wattage, they turn in a much slower time than when the wattages are unmatched, even though the stronger rider’s wattage is the same in both tests.
That’s not to say the matched wattage times are particularly slow, though. They just seem slow because the unmatched wattage test times are way too fast.
For example: a solo rider holding 300W climbs the Alpe in 49:28. But a banded group with one rider holding 300W and the other holding 75W climbs the Alpe in 40:11!
The other odd thing is how the pair times don’t change as the weaker rider’s power decreases. (This same phenomenon was observed in our flat tests.) That is, when Rider 1 is holding 300W and Rider 2 is holding 225W, they’re traveling at the same speed as when Rider 1 holds 300W and Rider 2 holds 75W. That’s strange. Shouldn’t groups travel significantly faster if the group’s average power increases significantly?
Test 2: Solo Rider
Next, I wanted to test how a solo rider in a rubberbanded Club event performs compared to the paired riders above. So I created an event with “Keep Together” enabled, but only had one rider join. Here are the results:
| Rider 1 | Time | Speed (kph) |
| 300W | 55:10 | 13.1 |
| 225W | 1:12:16 | 10.1 |
| 150W | 1:45:42 | 6.9 |
Compare those numbers above to free ride times with a solo rider (which should, in theory, be the same):
| Rider 1 | Time | Speed (kph) |
| 300W | 49:28 | 14.8 |
| 225W | 1:04:44 | 11.3 |
| 150W | 1:35:37 | 7.7 |
Observations
First, notice that when there is just one rider in the rubberbanded event, their speeds are much slower than they would be when free riding. That’s goofy. The speed difference between a solo rider in a banded event vs free riding isn’t as dramatic on the Alpe as it was in our flat tests, but there is still a significant difference.
Also, notice how the solo tests all came in significantly slower than the paired rider tests, even though logically that wouldn’t happen.
Concluding Thoughts
Simply put, Keep Together mode is broken up Alpe du Zwift. Zwift supposedly slowed banded group speeds on climbs in their version 1.52 update, but based on the results above, speeds are still too high!
I’d love to see rubberbanding improved so it’s a more realistic experience, but we also think Zwift should consider flagging all banded rides as “Gamified” when uploading to Strava, so we don’t set new segment PRs thanks to unrealistically high speeds in banded rides. Holoreplay and other in-game PR data also should not be saved for banded events, and times should not show on leaderboards outside of the event itself.
Coming Up Next
Soon, I’ll perform similar tests using banded Meetups, to see if/how Meetup banding behaves differently from Club Event banding. Stay tuned!
Questions or Comments?
Share below!
What is most disappointing is the flaw in banded rides on flat routes. All thanks to their attempt to slow down the band on climbs and we all know that failed.
Exactly my thought…..
Firstly, thanks for taking the time to test these scenario’s Eric!
Secondly, it still continues to amaze (and disappoint me) that, seemingly, no-one within Zwift is running these tests up front and spotting / fixing the errors.
Assuming you pass your findings back to Zwift directly… is it news to them when you give them the feedback or is there more testing that goes on internally and it simply never makes it to the top of the priority list?
Thankfully this feature is not one that I/we use often but it’s interesting to know! 👍
I don’t know what Zwift does internally to test this stuff. I only started testing these because I’d heard complaints of unrealistic speeds, and because I was curious how they actually worked.
That said, I’ve been in touch with a Zwift engineer just today who is fixing the issues discussed in these speed tests! So that’s good news. My guess is we’ll see the fix pretty soon.
Definitely want to see them flagged as gamified for Strava. I do enjoy an occasional banded ride, but hate my PR’s becoming out of my reach.
Agree this is broke. I did a ride as part of doddie aid earlier in the year and the time was something like 47 minutes – waaaaay beyond anything I’m capable of.
Of course I don’t have the stats for others on the ride to apply any science but i checked some other riders Strava activities. People like Mark Beaumont and Chris Hoy who I would imagine were some of the stronger riders on the event (and theoretically should therefore be slowed down by rubber banding); they also had PBs.
I self-reported in the end on Strava as I didn’t want to break my segment PBs.
Yeah, your experience is a good one pointing to how it’s broken. Slower/weaker riders should logically be getting PRs on banded rides… that makes sense. But if the stronger riders get a Strava PR too… well, we have a problem!
This is the ride if you’re interested in digging into it. 27 kmph 🤯
https://strava.app.link/1BSgU8Nq9Hb
will remove if this is against site rules!
I was on that ride too and wondered how valid those times were. I’ve flagged it too.
Thanks for your work. The banded rides [and the complex programing] are so flawed.
Well that is even more screwy than I had originally thought. I really don’t like the banded rides, and even avoid the group workouts for the very same reason. I know what my speeds are generally and how and effort feels at speeds for most of the Zwift routes. When I get into a banded event everything feels wrong to me, so I just don’t do them.
I completely agree banded rides should be marked somehow both in Zwift and Strava. This is also why I won’t do banded club events that involve the Alpe. A sub-60 time is well beyond my ability and I don’t want that specific badge unless I earn it outright. Meetups seem to behave more realistically, and I’m looking forward to your comparison. I’m such a data geek and an avid (err masochistic) fan of AdZ. I know I’m not alone in this sickness 😆