This research (it is a single piece of research written up in 4 different ways) simply concerns taking a rough piece of wood (a program that is almost correct) and sanding down the edges (fixing a small number of test cases). I didn't say "programs can't have algorithmic insight", I said randomly generating problems by "evolutionary" means will not contain insight by any other means than coincidence. The research you linked doesn't contradict that because all it concerns is smoothing down rough edges. Degeneracy is one of the fundamental features of a genetic code that is required for the theory of evolution to apply so I don't know why you say that "doesn't wash", it's a fact.