This work presents elements for an alternative operationalization of monitoring and diagnosis of multi-agent systems, developed in the context of compliance checking. In contrast to traditional accounts of model-based diagnosis, and most proposals concerning non-compliance, our method does not consider any commitment towards the individual unit of agency. Identity is considered to be mostly an attribute to assign responsibility, and not as the only referent to a source of intentionality. The proposed method requires as input a set of prototypical agent-roles (...) known to be relevant for the domain, and an observation, i.e. evidence collected by a monitor agent. We elaborate on a concrete example concerning tax frauds in real-estate transactions. (shrink)