How scouting a partner’s plans can favor cooperation

A coordination problem with a preparation stage

In the classic Stag Hunt game, two players choose between a cooperative, high-reward action and a safer alternative. Both do best when they cooperate. But choosing cooperation alone produces the lowest individual payoff, making the safer option attractive when a partner’s intentions are uncertain.

Wolfgang Kuhle’s arXiv preprint, Equilibrium Selection in Coordination Games with Planned Actions and Scouting, changes what happens before those choices take effect. Players prepare an action, receive an imperfect signal about their partner’s preparation, and then decide whether to proceed or request another attempt. The paper presents mathematical proofs, not experiments or an analysis of observed behavior.

Its central finding is conditional: increasingly accurate scouting and increasingly cheap revisions can make the best joint outcome overwhelmingly likely in every equilibrium, provided a specified kind of occasional cooperative mistake remains possible.

How the model works

The model has two players and two actions, labeled A and B. Mutual A gives both players their highest payoff. Mutual B gives both a lower payoff. When their choices differ, the A player receives the lowest payoff, while the B player receives more than under mutual B, but less than under mutual A.

Each attempt has three stages. First, both players simultaneously prepare A or B. Second, each privately receives a noisy indication of the other’s preparation. Signals are more likely to be correct than incorrect, and their accuracy can differ between players. Signals are independent across players and attempts.

Third, each player chooses whether to execute or revise. Execution requires both players’ consent. If either requests revision, both return to the preparation stage. Delay reduces the value of the eventual reward through discounting; endless revision yields zero. Cheap revision therefore means that waiting imposes little loss, not that delay is literally free.

The author studies Perfect Bayesian Equilibria: strategies and beliefs in which players update their information using Bayes’ rule and cannot improve their expected payoffs by deviating.

Why observation alone is insufficient

Without mistakes, both familiar coordinated outcomes remain equilibria. If a player knows the opponent always prepares B, a favorable scouting signal does not create a genuine prospect of cooperation: the player recognizes that the signal can be wrong. Mutual A also remains an equilibrium.

The protocol additionally permits a degenerate equilibrium in which both players always request revision and never collect a payoff. Thus, introducing visible preparations does not by itself guarantee successful coordination.

The selection mechanism needs occasional departures from intended behavior, called “trembles.” These create cooperative opportunities that a player can try to identify through scouting.

Screening for a favorable match

Consider a player who repeatedly prepares A but agrees to execution only after receiving a signal suggesting the opponent prepared A. With accurate signals, this strategy filters out most unfavorable matches. With inexpensive delay, waiting for a rare cooperative opportunity becomes more attractive.

The paper first develops this argument against strategies that remain unchanged across attempts. Its broader result allows strategies to depend on players’ private histories. For that result, each player independently has a fixed positive probability, in every attempt, of mechanically preparing A and consenting only after a signal indicating the opponent prepared A.

Under this particular perturbation, the author derives lower bounds on equilibrium payoffs and on the probability of eventual mutual A. As both signal accuracies approach certainty and the discount factor approaches one, those bounds approach the maximum payoff and probability one, respectively.

This is a limiting result with the tremble probability held fixed. It does not establish that arbitrary mistakes improve cooperation, or that realistic levels of monitoring and patience necessarily suffice. Nor does it require a unique equilibrium: the claim concerns outcomes across every equilibrium under the stated conditions.

The right to refuse is essential

The paper then allows a player to force execution despite the opponent’s request to revise. Costless forcing restores equilibria corresponding to the original game, including safe and mixed behavior when trembles are sufficiently small. Scouting cannot reliably protect a cooperative player who cannot refuse an unfavorable encounter.

A sufficiently high forcing cost restores the selection result. Some small positive costs still permit the safe equilibrium; the paper leaves a complete account of intermediate costs open.

The contribution is therefore a proposed theoretical mechanism, not evidence that monitoring reliably produces real-world cooperation. Its scope is this two-player coordination game, with independently renewed perturbations. Persistent behavioral types and extensions to games such as the prisoner’s dilemma remain open questions. The supplied work is an arXiv preprint, and its conclusions should be read as model-dependent mathematical claims.