Collective Perception (CP) allows connected autonomous vehicles to share and fuse processed sensor data via V2X communication. It can potentially allow for an increased object update rate, extended field-of-view awareness, and redundancy but it first requires a thorough evaluation and validation. Due to the CP’s field testing practical challenges most of the previous work on CP has considered large scale-simulations with a focus on connectivity/network aspects. More recently, large-scale collaborative perception synthetic datasets and open source benchmarks have appeared, allowing the perception engineers to study CP from a perception point of view, which is missing so far. In this paper, the first building blocks (work in progress) towards CP scenario-based testing for a roundabout navigation use case are been set by proposing a Bayesian CP algorithm and its testing plan. The CP algorithm is described and metrics for CP assessment are discussed focusing on the fused information content produced by the algorithm. The next steps towards a hybrid evaluation plan combining real-world agents and simulation are outlined.