Understanding how KOMs are verified starts with an uncomfortable fact: a segment result is accepted because nothing obviously contradicted it, not because anything confirmed it. There is no independent observer on the road. The platform receives a file and evaluates the file, which is a narrower thing than verifying a performance.
What the automatic segment checks actually test
Matching comes first. The track has to enter the segment near its start, follow roughly its shape, and leave near its end, within a tolerance that allows for ordinary positional error. A ride that cuts a corner or rejoins halfway does not match and never reaches the leaderboard at all.
Speed thresholds come next, tied to the declared activity type. A run at forty kilometres per hour and a bike ride at ninety both fail on their face. These catch the vehicle cases, which are the overwhelming majority of bad entries, and they catch them without anybody having to look at anything.
Where the checks stop
Nothing tests whether the athlete actually produced the power. Nothing tests whether the altitude was real, whether the file was edited before upload, or whether the rider was even on the declared bicycle. A file that is internally consistent and physically unremarkable passes, because internal consistency is the only thing being examined.
This is not an oversight so much as a hard boundary. A platform can audit its own data thoroughly. It cannot audit the world that data claims to describe, and no amount of server-side cleverness will ever close that gap. The verification is real, but it is verification of a recording.
The part that depends on people
Community reporting does most of the remaining work. Local riders know a climb, know roughly what a plausible time is, and notice when a result sits far outside it. That knowledge is not available to any automatic check, and it is why leaderboards stay usable despite the limits described above.
The weakness is that it scales extremely badly. A popular urban climb has hundreds of people who would notice an anomaly within a day of it appearing. A rural segment ridden twice a month has nobody, and an implausible result there can sit unchallenged for years without anything being wrong with the system.
How a disputed course record gets resolved
In practice, by evidence and very little else. A power file from a meter, a photograph with a timestamp, a second recording from a training partner or a coherent explanation of a device fault all move a dispute. Assertions about who seems fast do not, and neither does the volume of the complaint.
Where no evidence exists on either side, an arithmetic check is the neutral ground. Reconstructing the power a time demanded does not prove anything about the rider, but it establishes whether the result sits inside or outside recorded human performance, which is usually the question people are actually arguing about.
| Carries weight | Carries none |
|---|---|
| A power meter file for the effort | The time looking too fast |
| A second device recording the same ride | The rider being unknown locally |
| A demonstrable recording fault in the file | Consensus in the comments |
| Implied power outside the published range | How long the record has stood |
Why segment records drift upward over time
Leaderboards accumulate their errors instead of shedding them. A legitimate record is beaten and quietly replaced; an erroneous one stays until somebody actively removes it. Over several years this ratchet pushes the top of a leaderboard toward the fastest recording error the segment has ever produced, not the fastest ride.
The effect is strongest on short segments, where a brief position jump covers a meaningful fraction of the total distance. On a twenty-second sprint, two seconds of satellite confusion amounts to a tenth of the entire effort. On a forty-minute climb the identical error simply disappears completely into the noise.
KOM verification questions
Does anyone check a KOM before it appears?
Not in the sense of a human review. Matching and speed-threshold checks run automatically as the file is processed, and the result appears if it passes them. Any further scrutiny happens afterwards, and only because another athlete looked at the leaderboard and thought something was wrong. Nothing in the pipeline is waiting to assess whether a performance was humanly possible.
Can a power meter file prove a KOM was genuine?
It is the strongest evidence available, though not proof. A meter records what the drivetrain received, which is very hard to fake convincingly and correlates with everything else in the file. A coherent power trace consistent with the terrain settles most disputes. What it cannot do is show that the person who uploaded it produced it.
Why do implausible times survive on quiet segments?
Because the correcting mechanism is other people, and quiet segments have none. Automatic checks catch vehicles, not merely improbable performances. On a road ridden a few times a month there may be nobody with the local knowledge to recognise that a time is impossible. Quiet segments are therefore the least reliable leaderboards on the platform, not the most.
Does the platform remove records it considers suspicious?
It removes what fails its checks. Beyond that, a result that is merely remarkable is left alone, which is the correct default: judging performances rather than recordings would mean deciding who is capable of what, and that is not a decision a leaderboard can make. The line sits at what the file can be shown to contain.
Is a segment record on a short sprint less reliable?
Considerably. Positional error is roughly constant in metres, so it occupies a far larger share of a short segment than a long one. The shorter the segment, the more a momentary satellite glitch can distort the time, which is why sprint leaderboards are noisier than climbing ones. Treat any sub-thirty-second record as provisional until several riders have matched it.
Related: the plausibility check reconstructs the power a time demanded, flagged activities covers what happens after a report, and GPS accuracy explains the errors involved.