Run-time models have been proven beneficial in the past for predicting upcoming quality flaws in cloud applications. Observation approaches relate measurements to executed code whereas prediction models oriented towards design components are commonly applied to reflect reconfigurations in the cloud. Levels of abstraction differ between code observations and these prediction models. In this paper, the authors address the specification of causal relations between observation data and a component-based run-time prediction model. They introduce a meta-model for observation data, based on which they propose a mapping language to bridge divergent levels of abstraction and trigger model updates.