In the last decade, we have witnessed an increasing interest in reusable launchers. The clustered aerospike engine (i.e., a cluster of thrusters, referred to as cells in the following, that feed an aerospike nozzle) is an appealing technology for this application, for at least two reasons. First, an aerospike nozzle allows for continuous altitude adaptation during the atmospheric phases of flight, avoiding issues related to over-expansion that limit the design of conventional nozzles. Second, each cell can be throttled or cut off independently, enabling differential throttling. These features can be exploited twice, during both the ascent and the landing phases. In this context, the needed ability to predict engine performance, with sufficient accuracy and low computational effort, is challenging, especially considering plume interaction during differential throttling or possible failure modes.