Assuming that they ran the experiment in earnest, doesn't that strongly indicate that whatever code quality metrics in their product target are a pretty good signal? (Until people target those metrics)