First real test: a webcam, a laptop, and about forty people who had been told nothing beforehand. Here is what broke.
They did not know how hard to try
We asked the room to stand up and sit down to move an object on screen. Roughly half stood all the way, a quarter half-stood, and the rest stayed seated waiting to see what happened.
We had been designing for a binary — standing or not standing — and got a continuous, heavily-weighted-toward-nothing distribution instead. That is not a bug in the crowd. It is the correct response to an unclear instruction, and it meant our calibrated range was wrong within about four seconds.
The room found the exploit immediately
Within a minute someone worked out that waving both arms overhead moved the reading further than standing did, and told the people around them. The whole room switched to arms, which we had not calibrated for, and the game became unplayable.
Worth saying plainly: a crowd will find the cheapest input that moves the number, and will share it. That is not misbehaviour, it is the same optimisation any player does. The design has to assume it.
Forty people is not a small version of four hundred
With forty, individual behaviour dominates. One enthusiastic person visibly moves the aggregate, which breaks the central premise that no individual matters much. The cooperation moment — where people realise they need each other — never really arrived, because they did not need each other.
Our suspicion is there is a floor somewhere around a hundred people below which this genre does not work at all, and we should stop testing with small groups and calling it validation.
The webcam was the wrong tool
Mounted at laptop height, at the front, looking slightly up. The front row filled most of the frame and the back two rows were barely resolved, so the game was effectively being played by eight people while thirty-two watched.
Height is not a nice-to-have. We are treating it as the first requirement rather than a setup detail.
What actually worked
When it did respond, the reaction was exactly what we hoped for — people looked at each other, not at the screen, and started coordinating out loud without being asked. Thirty seconds of that was enough to convince us the idea is right even though the execution was not.