The model is calibrated based on the study from late march, which randomly sampled 700 individuals in Stockholm. From that they can infer the number of infected

The model is calibrated based on the study from late march, which randomly sampled 700 individuals in Stockholm. From that they can infer the number of infected people at that point in time. They use two separate variables to track unknown infections and known infections (hospitalizations), so from that study and hospital data they get a baseline. They can then verify how the model holds up for the group of hospitalized cases. There are charts towards the end of the paper which shows the real data plotted compared to the predictions, with confidence intervalls. The data is obviously very noisy, but it seems to track reasonably well, and is expected to taper of only slowly. The expected decline is not by any means drastic.