Tracking individual clouds in geostationary satellite observations and numerical simulations allows to quantify cloud lifetime and development. These higher-dimensional parameters are used in a novel approach to assess the fidelity of cloud-resolving models. Clouds are tracked in two months of satellite observations and accompanying model simulations of trade-wind cumulus clouds during the EUREC$^4$A campaign. Two ICON-LEM simulations with different spatial resolution give time series of cloud cover that reasonably resemble the GOES-16 observations, despite large differences in the number of clouds. Analysis of the tracked clouds yields similar distributions of cloud size, development, and lifetime in line with the expected values for different dynamic regimes from energy cascade theory. However, we find that the model underestimates the occurrence frequency of larger, longer-lived clouds. This effect is enhanced for finer model resolutions and might be related to the transfer of energy from larger-scale processes into the domain of the large-eddy simulation.