Following the "convention over configuration" paradigm, modeldriven software development (MDSD) generates code to implement the "default" behaviour that has been specified by a template separate from the input model, reducing the decision effort of developers. On the one hand, developers can produce end-products without a full understanding of the templates; on the other hand, the tacit knowledge in the templates is subtle to diagnose when a runtime software failure occurs. Using a concrete example, we discuss in this chapter the challenges of runtime problem analysis for MDSD, highlighting a potentially valuable research agenda for diagnosing the model-driven developed systems at runtime.