In this paper, we define a continuum of modeling styles, ranging from collections of very simple agents on one end to collections of very complex agents at the other end, and a simulation support mechanism that makes them all interoperable with ordinary discrete event simulation programs (at the simulation level; application-domain interoperability is harder). We show how all of these simulations have a stability question, and how some can be arranged to avoid it, by applying the mathematical theory of branching processes to study and even manage the stability of the system. This theory gives a predictive model of the density of simulation objects and events, according to a stochastic model that can either be matched to the observed execution properties of the simulation at run time, or built before the simulation is built, according to its expected local behavior.