Moving or stopped? Ask the position, not the speedometer
A phone's GPS speed reading lags and clamps. Three wrong explanations of a vehicle's behaviour came from trusting it; one short calculation on positions settled the question.
Chimes shows customers where an ice cream van is. When a van stops for a sale, its marker should settle where it actually is. To decide “the van has stopped”, the app samples the phone’s location and needs two consecutive stationary readings before it publishes the stopped position.
On a test drive, one of those readings looked wrong. The phone reported the vehicle as almost stationary while, a few seconds later, another reading said it was doing over 2 m/s. Which was right?
Three explanations, all wrong
We wrote three explanations, one after another, and retracted each:
- “The vehicle was doing 6.8 mph when the phone said it had stopped.” The 6.8 mph was a distance averaged across a minute between two samples. It described neither moment.
- “The one-off location request is unreliable; the continuous stream said ~7 m/s.” Every “disagreement” paired a reading with a stream sample 6–18 seconds away. The gaps produced the result.
- “One odd sample in nineteen; watch for a repeat.” The driver pointed out that the drive was stop-start. A short halt between two moving samples was not odd at all; it was likely.
All three treated the phone’s reported speed as the measurement. Position was sitting in the same log lines the whole time.
The calculation that settled it
Take consecutive positions, divide distance by time, and compare with what the phone reported:
| Interval | Gap | Distance | Implied speed | Reported speed | |
|---|---|---|---|---|---|
| 1 | 3 s | 22.1 m | 7.37 m/s | 6.95 m/s | approaching |
| 2 | 3 s | 20.0 m | 6.65 m/s | 5.36 m/s | braking |
| 3 | 9 s | 18.9 m | 2.10 m/s | 0.13 m/s | stopping |
| 4 | 10 s | 1.9 m | 0.19 m/s | 2.09 m/s | stationary |
| 5 | 4 s | 21.8 m | 5.45 m/s | 5.64 m/s | pulling away |
1.9 metres in 10 seconds: the vehicle was stopped. The reading we had doubted (0.13 m/s) was right. The suspect value was the 2.09 m/s, reported while the car had barely moved: phone GPS speed lags position when a vehicle accelerates.
The app had behaved exactly as designed. It confirmed two genuine stops and published the second.
What we keep
- Speed from a phone is derived. It lags on acceleration, clamps to zero below a confidence floor, and reports -1 when unknown. Position is what the tracker publishes and what the customer sees, so position is the ground truth.
- When explanations keep getting narrower, test the premise they share. All three versions assumed the speed channel was the measurement.
- Ask the person who was there. One sentence from the driver did what three rounds of analysis could not.
This is the kind of movement data that arrival prediction, in our Sense research, is built on: we only trust a model as far as we trust the signal it learns from.