Which recording is fit for corner analysis?

Kurvenfokus imports rides from nine sources — from a calimoto GPX to a RaceBox at 25 Hz. We measured what the files actually contain: how often they store the position, where the gaps are, whether timestamps and lean angle come along. The result is surprising: for corner counting, the best sensors do not make the best source.

1 · The sources at a glance

What each source delivers and how it gets into Kurvenfokus. Sampling rates for calimoto, Motobit, Komoot, Kurviger, RideLink GPX and RaceBox measured on September 15, 2026 across 212 rides; Relive from two files of April 2026, Curve Tracker from the cross-check of September 22, 2026, MotionRecord from the product documentation. Fields from the export formats.

¹ Komoot: the September file samples at 1.00 Hz; an older file from April sat at 0.2 Hz with gaps up to 63 s — the export apparently depends on the setting. “Computed” means: Kurvenfokus derives the lean angle from speed and corner radius, as described on the technology page. RaceBox VBO is currently read by the iOS app only. Relive and Kurviger go through the generic GPX import.

2 · GPS geometry: 212 rides, 14 sources

Evaluation of September 15, 2026 across our entire GPX and VBO data. Point density in Python, corner counting with the production code of Kurvenfokus iOS.

The median point spacing measures riding speed (speed ÷ sampling rate), not quality — see the addendum in section 5. Gaps are real jumps over 100 m; 1–8 per 100 km in recordings, 195–313 in planned routes, because planners only place a point at direction changes. ² Via the GPX detour our pipeline counted 0 corners in this test — the cause was our source detection, not the file; the current version recognizes RaceBox GPX. The intended path is the VBO import anyway. ³ Not counted without timestamps (Curve Tracker GPX) or in the WingMan JSON. The two Curve Tracker rows are two export states of the same program from two folders; for the first group the pipeline delivered no corner count. Corners per km are comparable only within the same source.

3 · What this means for corner detection

Three rules that follow from the measurements — and that Kurvenfokus follows in the app and on this website.

1. 1

   Compare corner counts only within the same source

   Kurvenfokus detects corners from the shape of the track. Denser recording means more detected corners on the same road. A planned calimoto route has 361 m point spacing, the ridden recording 14 m — which is why a different number shows up there. The example is worked through on the page 1,997 corners (German).

2. 2

   Without naming the device, a number is accurate to about 5%

   The same ride, iPhone 11 and iPhone 12 at once: 85.72 vs. 90.24 km unfiltered. Where the source supplies a GPS accuracy, Kurvenfokus discards poor points — from 25 m for MotionRecord imports, from 50 m for the other JSON sources. In our measurement a 25 m threshold shrank the gap between the two devices to 0.07 km, discarding 0.2% of the points. GPX files carry no accuracy value; no filter applies there.

3. 3

   GPX without timestamps: speed and lean angle are estimates

   If the file has no time, every evaluation has to assume the speed — and with it the lean angle computed from it. This affects the Curve Tracker GPX export and some converter outputs. Better: import the original format (storage.json, RideLink JSON, VBO).

4 · Curve Tracker in detail

The app stores more than its GPX export reveals. What arrives by which path — and why Kurvenfokus does not take over the sensor angles from the app database.

5 · Data basis: which data, when, how often

This page rests on four measurement series and one archive. Every number above can be traced to a measurement here.

Population: our ride archive (as of September 28, 2026)

The samples come from our own archive with 688 rides, 61,592 km, August 2019 to September 2026. The figures come from the catalog via SQL, not from estimates.

Motobit, Komoot, Relive and Kurviger exist only as individual GPX files (evaluation of September 15), not in the archive. RideLink: 147 of the 570 recordings are file variants of the same ride and count only once for rides, kilometers and GPS points. Kilometers per source via the assigned rides; six rides (770 km) were recorded by several devices at once, 2 of them by three; they count for each source, which adds 918 km of multiple counting — that is why the column adds up to 62,510 km instead of 61,592 km. Sampling rates come from the interval between consecutive points — not from the recording duration, which includes breaks. Method: our own recordings and rides from test riders with their consent; evaluation with AI assistance, cross-checks by hand. No lab measurement, no manufacturer commission.

Sensors in detail

This page rates the sources as corner sources. If you want to know which app delivers which sensors at which rate — acceleration, gyroscope, barometer, signal processing — the technical test report with star ratings and field lists is on motionrecord.de.

calimoto, Motobit, Komoot, Kurviger, Relive, RideLink, WingMan, Curve Tracker, MotoLeaning, RaceBox, BMW Connected and Kurvenheld are trademarks or products of their respective owners and are not affiliated with Kurvenfokus or MVXLabs. They are named solely to state format compatibility and to describe our own recordings. All information without guarantee, as of September 28, 2026.