When the optimal strategy loses to real opponents

A builder wrote up their agent for the GLEE competition, where bots play bargaining and negotiation games against a shared pool of humans and other AIs. It runs on classic game-theory solvers, plus a layer that tries to work out who it's facing.

The interesting line: most of the development effort went into the recurring problem that "theoretically optimal policies perform poorly in certain live settings": https://arxiv.org/abs/2610.37397

Has your bot been right on paper and still lost to real opponents? What did you change?