The student who tested the counter

Three weeks after the first version appeared, a student in grade nine sent a message to the team: your counter is wrong. She had done twenty squats and the screen showed seventeen.

The easy response would have been to explain why she was mistaken. Instead the team asked her to do something more useful: repeat the test while a classmate counted aloud, and film the screen, not the person. The result confirmed her claim. Three repetitions in the middle of the set were not counted.

Finding the cause took two evenings. The movements that failed were faster than the others, and the rule required the body to stay below a certain angle for a minimum time. The minimum had been chosen to stop cheating, with a value that seemed sensible and had never been tested against real students. For a quick, correct squat it was slightly too long.

The fix was to measure the time between the lowest point and the return, rather than the time spent below the angle. The new rule still rejects a bounce, because a bounce does not reach the angle at all. It was tested against recorded data from forty sessions before it was released.

The interesting part is not the bug. It is that the person who found it was not on the team, had no access to the code, and was taken seriously. A system that treats such a message as a complaint loses the information. A system that treats it as a test gains a tester.