Operationalizing compatibility with strategy-stealing — AI Alignment Forum