Security designation tools are a key feature of all prisons in the United States, intended as objective measures of risk that funnel inmates into security levels-to prison environments varying in degree of intrusiveness, restriction, dangerousness, and cost. These tools are mostly (if not all) validated by measuring inmates on a set of characteristics, using scores from summations of that information to assign inmates to prisons of varying security level, and then observing whether inmates assumed more risky did in fact offend more. That approach leaves open the possibility of endogeneity--that the harsher prisons are themselves bringing about higher misconduct and thus biasing coefficients assessing individual risk. The current study assesses this potential bias by following an entry cohort of inmates to more than 100 facilities in the Federal Bureau of Prisons (BOP) and exploiting the substantial variation in classification scores within a given prison that derive from systematic overrides of security-level designations for reasons not associated with risk of misconduct. By estimating pooled models of misconduct along with prison-fixed effects specifications, the data show that a portion of the predictive accuracy thought associated with the risk-designation tool used in BOP was a function of facility-level contamination (endogeneity).