We use administrative register data from the Norwegian child welfare service (CWS) to analyse the outcomes of 252 144 CWS referrals received between 2014 and 2020. Using a Bayesian hierarchical model, we document extensive and systematic variation across time, across referral sources and across referral concerns. Defining high severity cases as cases in which the CWS seeks mandatory interventions, mandated and nonmandated reporters have similar substantiation rates, but this shared probability declines by half across our data period. The differences in substantiation rates across reporting sources persist when comparing referrals with identical referral concerns. Drawing on the extant literature, we suggest that occupational groups may have evolved diverging reporting practices as a consequence of vague official guidelines combined with a lack of systematic training in identifying relevant cues of maltreatment. We also highlight a possible tension in the Norwegian system: while the family-oriented Norwegian CWS largely opens cases based on referrals, its aim is to work preventively and voluntarily with families before issues reach the severity covered by existing reporting mandates.