Precision neurorehabilitation is attracting increasing attention to ensure careful targeting of rehabilitation intervention to optimize individual patient outcomes. However, this requires models that support interventional reasoning about therapy ingredients, going beyond purely associational models and biomarkers. Developments in the field of causal inference can be leveraged to build such models with explicit mechanistic specifications. Focusing on upper-limb robot-assisted therapy, this work presents a structural causal model in the form of a directed acyclic graph (DAG) synthesising our current understanding of robot-assisted neurorehabilitation. The DAG captures the known and hypothesised key constructs and their causal influences as nodes and directed edges, respectively, reflecting the current domain knowledge in robot-assisted upper-limb neurorehabilitation. This theoretical work details the nature of the different components of the proposed model, along with how a fully specified causal model could be used for making optimal therapy planning and design. We anticipate the proposed model will serve a catalytic role in advancing our mechanistic understanding of robot-assisted therapy, and help develop precision neurorehabilitation approaches for optimal individual patient outcomes.