We model stable matchings under group-dependent bias and correlated evaluations, characterize equilibrium thresholds, and show how evaluator alignment amplifies or mitigates fairness loss in decentralized systems.