The averaging period doesn't last the whole game: "we set a maximum of 600 APMs over 5 second periods, 400 over 15 second periods, 320 over 30 second periods, and 300 over 60 second period. If the agent issues more actions in such periods, we drop / ignore the actions. These were values taken from human statistics." [1]

I'd be curious about the approach they used to draw from human statistics. It's possible that peaks in the human APM distribution are caused by e.g. holding down keys, resulting in ~2000 APM [2].

[1] https://www.reddit.com/r/MachineLearning/comments/ajgzoc/we_...

[2] https://blog.usejournal.com/an-analysis-on-how-deepminds-sta...