If I understand correctly, the paper asks us to imagine Alice's friend making a 'classical' coin flip, and Alice modeling and measuring her friend's lab as if the coin flip were a quantum superposition of coin heads/coin tails.

But this is simply wrong. The error creeps in in the second assumption:

> An agent can analyze another system, even a complex one including other agents, using quantum mechanics

This is likely a good assumption, but quantum physics would tell Alice that the coin's atoms are all firmly in a state corresponding to either 'heads' or 'tails'. The author is actually making the following assumption:

> What we classically think of as chance is completely interchangeable with the unpredictability of quantum superpositions.

This notion has already been proven wrong many times over.