Mental health service systems are inherently complex, both in their detail and dynamics. System dynamics offers great potential to help policy makers, administrators, and researchers make better decisions about service system changes. However, efforts have been constrained by not being able to construct numerical reference modes without making strong assumptions about the structure of the case flows. This paper presents a novel approach to generating numerical reference modes from administrative databases that is based on computation theory. The method is validated with simulated datasets, and its feasibility and substantive significance demonstrated in an analysis of a merged child welfare database containing 10,250 children and adolescents.