Runtime Verification (RV) involves monitoring a system to check if it satisfies or violates a property. It is effective at bridging the reality gap between design-time assumptions and run-time environments; which is especially useful for robotic systems, because they operate in the real-world. This paper presents an RV approach that uses a Communicating Sequential Processes (CSP) model, derived from natural-language safety documents, as a runtime monitor. We describe our modelling process and monitoring toolchain, Varanus. The approach is demonstrated on a teleoperated robotic system, called MASCOT, which enables remote operations inside a nuclear reactor. We show how the safety design documents for the MASCOT system were modelled (including how modelling revealed an underspecification in the document) and evaluate the utility of the Varanus toolchain. As far as we know, this is the first RV approach to directly use a CSP model. This approach provides traceability of the safety properties from the documentation to the system, a verified monitor for RV, and validation of the safety documents themselves.