GPS accuracy is usually quoted as a single number, which hides how the error behaves. A receiver solves for position by timing signals from satellites whose positions it knows. Anything that delays a signal or blocks it shifts the solution, and the shift is rarely random in the way averaging would fix.
Multipath: the error that matters most to cyclists
A signal that bounces off a building or a rock face arrives later than one travelling directly. The receiver cannot tell the difference and treats the delay as extra distance, placing itself away from where it actually is. This is multipath error, and it is the dominant fault in real riding.
It explains why a trace through a city centre can run straight through the middle of buildings, and why a descent in a steep valley drifts sideways across a hillside. The receiver is not confused about which road you took. It has simply been given consistently wrong distances by reflected signals.
Why tree cover is different from a canyon
Foliage attenuates the signal, it does not reflect it. Signals still arrive but weakly, so the receiver drops satellites and solves with fewer, degrading the geometry. The result is a noisier trace that wanders around the true path, not a confident trace displaced to one side. Both hurt, but they leave different signatures.
What GPS accuracy costs a segment time
Error in metres stays roughly constant regardless of segment length, so its effect scales inversely with distance. Fifteen metres of displacement at the start of a twenty-second sprint is a large fraction of the whole effort. The same fifteen metres on a ten-kilometre climb is not detectable in the result at all.
This produces a systematic bias in leaderboards that nobody designed and nobody intended. Short segments collect more anomalous times than long ones simply because they offer the same fixed error far more leverage, which is worth remembering carefully before concluding that a particular sprint record must have been deliberately manipulated.
| Condition | Dominant error | Effect on the trace |
|---|---|---|
| Open road, clear sky | Receiver noise | A few metres of scatter |
| Dense woodland | Attenuation, lost satellites | Wandering, noisy path |
| City streets | Multipath reflection | Confident displacement, wrong side of the road |
| Steep valley | Blocked sky, multipath | Sideways drift across the slope |
| Tunnel | Total signal loss | Straight line between entry and exit |
Settings that help, and ones that do not
Recording rate is the one setting genuinely worth changing. One-second recording captures corners and short efforts properly; smart recording saves battery by dropping points, and it rounds off exactly the features a short segment depends on. For anything remotely competitive, the extra battery cost is an easy trade to make.
Enabling additional satellite constellations genuinely helps, because more visible satellites mean better geometry and more redundancy when some of them are blocked. Waiting for a proper fix before starting helps just as much, and costs nothing but patience. Most bad traces begin with a ride started thirty seconds too early.
What does not help is treating the accuracy figure on the screen as any kind of verdict. It describes the receiver's confidence in its own solution, and a receiver deceived by reflections is confidently wrong. A clean-looking accuracy number is entirely compatible with a trace that runs through a wall.
GPS accuracy questions
Why does my ride show me on the wrong side of the road?
Multipath, almost certainly. Signals reflecting off buildings or terrain arrive late, the receiver reads the delay as extra distance, and the computed position shifts away from the reflector. The trace looks confident because the receiver genuinely believes it, which is what makes the error hard to notice. The accuracy figure on screen will often look perfectly healthy while it happens.
Is a phone less accurate than a dedicated bike computer?
Often, though less than it once was. Phones use smaller antennas in worse positions, frequently in a pocket with a body between them and the sky. A bar-mounted unit with a clear view of the sky has a real advantage that has nothing to do with the quality of the chip inside.
Does one-second recording actually change segment times?
On short segments, yes. Smart recording drops points during steady riding and infers the path between those kept, which rounds corners and can shift where a segment is judged to start and end. If a short segment matters, switch to one-second recording before you ride it, not afterwards when nothing can be recovered.
Why does my trace go straight through a tunnel?
Because no signal reaches the receiver inside it. The device records the last good fix at the entrance and the next at the exit, and any line drawn between them is an interpolation. Distance through the tunnel is underestimated, and altitude recorded there is meaningless. Any segment beginning or ending inside one produces a time nobody should take seriously.
Can GPS errors be corrected after the ride?
Partly. Obvious jumps can be identified and cropped, and map-matching can snap a trace to known roads. Neither recovers information the receiver never had, so a badly degraded recording stays degraded. Identifying which points are affected is realistic; repairing them is not. That is the honest limit of every repair tool, including the audit published on this site.
Related: the GPX audit counts position jumps in your own file, elevation errors covers the vertical equivalent, and KOM verification explains what this means for leaderboards.