In the 20th century, nearly 66% of governments killed around 262,000,000 civilians. Existing research tends to explain the government's decision to kill by focusing on the conditions and incentives that surround murderous campaigns. Yet even when conditions and incentives to kill are similar, civilian body counts vary widely. I leverage the puzzle of variance in civilian death tolls by shifting the focus away from the decision to kill. Instead, I conceptualize killing as the outcome of a process of strategic interaction between the government that sends the killing order and the perpetrators who implement that order. A game-theoretic model reveals that under changing conditions, perpetrators may kill as many civilians as they are asked to kill, or they may kill any other number, including zero. Empirical tests support implications from the formal model, suggesting in part that international actors are well-equipped to protect civilian life. The solution to the puzzle is this: civilian death tolls may vary even when conditions and incentives to kill are similar, because the actors care not only about the killing that results from their interaction but also about the consequences of that killing for their own lives and liberties.