I have decided to shelve my work on architectures that are biologically inspired for now. I was getting reasonable results with spiking neural networks and evolutionary training, but there are so many hyper parameters to think about and how they behave over time is really hard to predict. I was also struggling deeply with how to manage topological concerns like network growth over time.
With interpreted evolutionary programs, the memory access patterns are so much more ideal with the program counter stepping through (mostly) contiguous bytes vs totally insane recurrent spiking neural access patterns. You get so many more generations & candidates evaluated per unit time that it can make previously apparent "dead ends" viable, simply because you don't need to have extreme patience to find out anymore. I am discovering that iteration speed is the most important thing in this arena. The faster you find out how bad a certain parameter adjustment is, the sooner you can get to the good ones.
I am also working on an unrelated contract to integrate some back office banking systems. Not much worth discussing there.