Many large-scale Internet applications optimize their overlay network to reduce latencies. Embedding coordinate systems like Vivaldi are valuable tools for this new range of applications since they propose light-weight algorithms that permit to estimate the latency between any pair of nodes without having to contact them first. It has been recently demonstrated that coordinate systems in general and Vivaldi in particular are sensible to attacks. Typically, nodes can lie about their coordinate and distort the coordinate space. In this paper, we propose a formal reputation model to detect misbehaving nodes and propose a reputation adaptation of Vivaldi called RVivaldi. We evaluate the performance of RVivaldi using the King dataset and show that RVivaldi is less sensitive to malicious nodes than Vivaldi.