Both classic and modern theories of rebel warfare emphasize the role of unexpected attacks against better equipped and larger government forces. We test empirical implications of a simple model of combat and information-gathering using highly detailed information about Afghan rebel attacks, military base infiltration, insurgent-led spy networks, and counterinsurgent operations. As rebels gather more resources, their at- tacks become temporally concentrated: a one standard deviation increase in opium revenue leads to a .3 standard deviation increase in temporal clustering of rebel at- tacks. In contrast, following abnormal battlefield losses (labor scarcity), the timing of insurgent attacks becomes less concentrated: a one standard deviation increase in labor scarcity increases randomization of attack timing by .12 standard deviations. We supplement our benchmark specification with a novel instrumental variables (IV) approach that uses high resolution data on agronomic inputs and dimensionality reduction to instrument for opium suitability. We use LASSO and sample randomization tests to assess and confirm the validity of our IV approach. The main effect is significantly enhanced in areas where rebels have the capacity to spy on and infiltrate military installations. We use proprietary military surveys to estimate exposure to informal taxation by government officials, which shows that relatively lower reservation wages lead to larger revenue effects. We find evidence that rebels exhaust their resources during the fighting season after taxation.