An evolutionary step happens on request; if there is an individual in the population with an unassigned fitness, it is chosen and passed to the evolved component. Otherwise, two parents are chosen with a probability proportional to their fitness. These are crossed over and mutated to form two offspring, which are reinserted into the population.
When individuals need to be removed from the population, they are selected probabilistically according to the inverse fitness and the similarity with other sequences. As for classifier systems, the idea is to enforce diversity in the population by removing the worst behavior when two are similar.