Analyses of compositional coevolution indicate collaboration methods seriously impact the optimization goals of the methods.   Experience suggests compositional coevolutionary algorithms that use averaging mechanisms for collaboration (e.g., complete mixing) perform well in multiagent learning settings, where the goal is often to find team behaviors that are not necessarily optimal, but robust to certain types of changes.  At the 2006 Dagstuhl seminar, I introduced a new framework for rigorously defining robustness.  In this year's talk, I extend this idea by introducing the notion of a parametric degree of robustness, and I demonstrate a particular condition that ensures progress in a dynamical systems model of the algorithm.  The condition relates this parametric degree of robustness with the relative sizes of portions of the phase space of the modeled algorithm.  In this way, one should be able to show that the larger the degree of robustness, the larger the basin of attraction for a particular solution:  robust solutions are attractive to certain kinds of CCEAs.