Beating the Pool, Not the Game: A White-Box Equilibrium Agent for Language-Based Economic Games

We describe sudsho, a fully rule-based agent for the GLEE Competition that plays bargaining, negotiation, and persuasion with zero inference-time LLM calls. Each game is scored as a payoff percentile against an empirical pool of prior play on the identical configuration, not against a rational opponent, so equilibrium analysis enters only as a bound on what to accept while demands are clamped toward the pool's empirical behavior. The agent is white-box by construction: every move is a deterministic function of the visible state (persuasion mixing aside), so alignment is verified by replaying the season's post-freeze move log through the policy. Over 117,280 games the agent reached an overall rating of 2056 at \$0 inference cost. The account, which is ranked by its best same-policy replica, placed 26th of 71 ranked accounts.