In a comment on my post about the book Compositional Evolution, William Bradford pointed out that the compositional operators which are supposed to enable greater evolutionary change are themselves embedded in and dependent upon complex cellular contexts. So, by themselves, they do not explain the problem of biological complexity.

This immediately suggests a few research problems. It seems that the problem of the origin of compositional variation methods must be described by a gradualist framework (unless we assume that the first reproducing organisms contained them, which seems unlikely). So, characterizations of these mechanisms as strongly interdependent and irreducibly complex would provide a greater problem to evolution than evolution with them does. This would be hard to do, because such mechanisms aren't easy to separate from the function of the rest of the cell. But then, perhaps that strengthens the point.

From a more formal perspective, can random-mutation genetic algorithms evolve different modes of variation which include compositional mechanisms? My hunch is that they can, but only by specifying the target ahead of time, that is front-loading the algorithm with everything it needs to know to reach the desired goal. Such a teleological search is not biologically plausible under Darwinian assumptions.