abstract = "Prior work using the Buffet method showed that Markov
brains will not only evolve to specialize on different
tasks, but when provided with different logical
components, will tend to create solutions that favor
the data processing units best suited for each task. we
replicate these results using new implementations (a
true replication) and conduct a more extensive
analysis. We consider not only component usage in
aggregate but also look at gate usage in individual
solutions. We find that hybrid solutions are common.
However, different tasks tend to use different
component types, and each task results in unique rates
of component mixing.",