Year: 2026
Type: preprint
Abstract: A memory-one commitment that controls stationary payoffs need not impose the same relation on rewards accumulated before an irreversible collision. We examine this distinction in repeated Chicken with payoff ordering í µí± > í µí± > í µí± > í µí±. Starting from cooperation balance, we specify a baseline-í µí± zero-determinant family and its exact probability constraints. A supporting payoff inequality establishes an attainable cooperative best response for a stated subset of Chicken games. We then apply finite cooperation-flow telescoping to encounters that stop at the first realised mutual defection. The resulting cumulative-payoff identity retains an endpoint term; independent execution errors add a cooperation-imbalance correction. Excluding only the collision payoffs differs from excluding the entire affine reward increment when the baseline is nonzero. Bounded horizons require no limiting assumption, while an unbounded version follows under finite expected collision time. Deterministic absorbing-process calculations use one main payoff tuple, (í µí±,í µí± ,í µí±,í µí±) = (5,3,1,0), for both continuing and stopped comparisons. Against a specified opponent at execution-error probability 0.02, a slope-two policy exceeds a unit-slope policy’s stationary X payoff by 0.7477, but its stopped cumulative payoff over 50 offered rounds is lower by 2.0210. Changing the opening-memory seed reverses this stopped comparison. The analysis distinguishes stationary response incentives from collision-terminated performance without asserting learning convergence or optimality of the stopped policies.
Source: Research Square
Author: M’Hamed CHERIF
Institution: Université Libre de Bruxelles
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Keywords: zero-determinant strategies, execution errors
Open Access status: green
OpenAlex ID: W7225903879
