Strava releases further details about HOW they protect leaderboard integrity.

Heinz famously has 57 varieties. That’s certainly helped the company do well. Strava has taken a bean out of the same can and last week told us they had 57 ways to remove anomalies from segment leaderboard results.

Every activity uploaded to Strava is now automatically analysed by our machine learning model, which looks at 57 different factors, like speed and acceleration, to determine if any portion of an activity was recorded in a vehicle. If the system detects a vehicle, the activity is flagged, and the user is prompted to crop it or make it private. Next we’re applying this same technology to prevent bike rides on run leaderboards and ebikes on ride leaderboards. This is part of our ongoing effort to keep leaderboards fair and reflect true performances. [via: Someone at Strava]

I was intrigued. 57 tests are a lot, and I asked the company to reveal their highly confidential internal algorithm. Naturally, they declined but were prepared to add some meat to the original statement.

We can share are the following factors, which are analysed by the machine model learning to identify activities uploading in a vehicle:

• Top speed
• Average speed
• Variance of acceleration
• Average acceleration

[via: someone else at Strava]

This doesn’t really help our understanding, and hence trust, about what is going on with the leaderboards. Clearly, Strava is checking that no one is averaging 100km/h up the Tourmalet. Still, it must be hard to distinguish the times when we accidentally leave a bike computer recording from people who set out to cheat and get KOMs/QOMs in a car by going just faster than the previous attempt. Reading between the lines, Strava seems to be trying to discern some of these anomalies by looking at the acceleration behaviour.

Anyway, the news last week was that the old algorithms that detected cars were now extended to detect misrecorded activities – like a bike ride recorded as a run. We’ve all accidentally done it. At least I have.

Each activity is automatically analyzed to ensure it is eligible for segments:

We pass all activities through a machine learning (“ML”) algorithm to check for the common mistakes that disrupt leaderboards. For example, uploading activities as the wrong sport type (like selecting “run” when you actually went for a ride or selecting “ride” when you were riding an e-bike) can cause inaccuracies on segment leaderboards. Another example includes leaving a GPS running while driving home after an activity, which can also disrupt leaderboards. To check for these and other issues, we pass all activities through an ML algorithm designed to catch these mistakes. If the model detects that something looks like a mistake during a portion of your activity, you’ll see a red banner at the top of your activity page [via: Segment Leaderboard guidelines, 2025]

Take Out

I find there are just too many segments on Strava. There are way too many duplicates, or near-duplicates.

In my opinion, Strava should determine massively more verified segments and be done with it. Those are the only segments we all see. If we want private or club segments, we can, but they won’t spoil the segment experience for everyone else. Maybe I’m just grumpy because I live in a highly cycled area and can’t get onto any of the proper leaderboards, although I manage it for the age-related ones.

While that would be a significant undertaking for Strava to clean up and verify all public segments, it would massively reduce the behind-the-scenes calculations and processes the company has to do each time we upload a ride.

On another realted point, there must be scope for Zwift, Strava and some other platforms to get together and produce a ‘verified rider’ status. Simply, that would be verification that you are real and appear not to cheat, perhaps even holding a simplified private power profile that the other platforms have access to aid in their verification efforts. I would love a check mark next to my real-world Strava name as a verified rider…even more so on the rare occasions when I try to beat my mates on Zwift but suspect they have lowered their weights too much.

Thoughts welcomed.