How accurate is it, and how do we know

Method

• Paired runs. Since May 2026, more than 35 runs by two runners have been recorded on both watches. Each Garmin FIT file and each RunToMax export is kept and compared: total distance, per-kilometre splits, heart rate, cadence and elevation.
• A scored corpus. Seventeen of those runs form a fixed test set with a written pass/fail gate: total distance within 1% of the reference, and per-kilometre splits whose error does not drift with speed. Any change to the distance code is replayed against all seventeen before it ships.
• An honest reference. A Garmin is not ground truth. On stop-start runs its headline distance can exceed the distance implied by its own speed record; in those cases we judge against the Garmin’s own integral, not its headline.

Distance — what ships today

RunToMax uses Apple’s workout distance engine — the same distance the Watch writes to Apple Health — fed by the Watch GPS and motion sensors. Results against the Garmin:

• Steady runs: within about 1%. Examples from the paired set: −0.93% on a 9.9 km run, −0.95% on a 9.0 km run, and +0.006% on a 7 km run — the closest agreement recorded.
• Interval and stop-heavy runs: 1–3% short of the Garmin. A 4 × 800 m session read 142 m short over 8.5 km (−1.7%); a run with a 95-second walk break read −2.8%, of which most was the two watches counting the stopped time differently.

Distance — what is in testing

A GPS filter that sums position deltas over a smoothed track runs in the shadow on every run and is scored against the corpus. On the most recent paired runs its totals are within ±0.3% of the Garmin, including runs with 25–33 stops. It is not shipped: per-kilometre splits on stop-dense kilometres still miss the 8-second bound on some runs, and two of three runs by the second runner showed speed-correlated drift. It ships when the gate passes, not before.

Splits

On clean runs, per-kilometre splits agree with the Garmin within about 5 seconds. Where they differ it is usually the kilometre that contains several stops: each pause and resume introduces roughly 1.5 seconds of clock disagreement between the two watches, and a kilometre with four stops can differ by 8–10 seconds. Lap pace shown live on the Watch is the same number the summary shows afterwards.

Cadence, heart rate, elevation

• Cadence from the Watch’s motion sensors matches the Garmin: median 180 vs 181 steps per minute across a 1,700-sample run.
• Heart rate is the Watch’s optical sensor. Two wrist sensors disagree run to run in both directions; we do not claim either is right.
• Elevation uses the barometer with GPS drift correction. Treat totals as accurate to about 9 metres per run.

Known limitations

• Two watches never agree exactly. Expect ±1% on any single run, from either device.
• Stop-heavy sessions are where the remaining error lives — pause handling, not GPS quality.
• Tall buildings, tunnels and dense tree cover degrade Watch GPS as they do any GPS.
• Treadmill distance is Apple’s stride estimate unless a pod provides it.
• Ground contact time from the wrist is not reliable enough to coach from, so RunToMax does not use it for form guidance.

If you record paired runs and see something different, send the files — the corpus grows one run at a time. Contact is on the support page.