Abstract: The inflammatory effect of large-scale indiscriminate violence has become an article of faith among practitioners and scholars of irregular war. Yet such practices frequently occur, often devoid of any discernible coercive logic. Why? Using a mathematical model of punishment and cooperation, I show that successful coercion is difficult to achieve in irregular conflicts, particularly for governments fighting guerrilla opponents. The identification problem -- the inability to correctly locate and punish one's opponents -- compels the side with an informational disadvantage to gain coercive leverage by escalating violence. Contrary to conventional wisdom, this diminished ``power to hurt'' neither moderates the intensity of violence in equilibrium, nor assures defeat in continued fighting. Rather, the identification problem creates powerful incentives to escalate -- particularly where a combatant cannot offset this disadvantage by offering generous rewards to potential supporters. I evaluate these propositions empirically using micro-level conflict data from Russia's North Caucasus.