I thought just the same. At first I thought there was an error with the counting, but there's a debounce of 0.2s for every pulse https://github.com/jlhonora/iot/blob/master/arduino/IoT/pir..... The wheel is not perfectly symmetric, and I'm placing the magnet on the opposite side of the reed switch used to count each lap, so that there's no risk of getting the switch stuck generating lots of pulses.

The activity graph (shown on the twitter feed) is for that same night, which seems to be quite consistent over time. There are no peaks either, so I think it should be accurate. After all, he does run all night.