It is difficult to achieve assured conformance between architecture and code. We lacked proper architecture descriptions in the context of the evolution of an industrial software product line of logistic systems. As a result, a lack of explicit architecture documentation about the deployed products led to ad hoc update practices, which were error prone and resulted in unnecessary and undesirable shutdowns. To tackle these problems, we have codified the architecture knowledge required for evolving products in a viewpoint. Based on this viewpoint, we developed a supporting tool that allows generating on-demand architecture models. These models guide maintainers by listing the concrete tasks they need to perform when upgrading a system and by showing inconsistencies when they fail to do so. The evaluation of 68 updates of industrial logistic systems demonstrates a significant improvement in the quality of system updates with respect to the correct execution of updates and a reduced interruption of services.