I have been playing around with per BP mutation rates encoded in the DNA as a means of simulating the degenerate nature of biological DNA codons. Basically, the idea is to make each and every individual location in a bot's genome - every sysvar, operator, number, flow control statement, etc. degenerate in that there are multiple underlying representations for each, each with say, 64 different levels of mutation probabilities from super stable to super high mutation propbability. These mutation probabilities would be visable to selection.
The idea is to maintain say, a vector of 8 high order bits alongside each BP which would influence the mutation probability at that location in the genome and to store that with the DNA, syntax TBD. Right now I'm just focused on the point mutation at that loci, but the idea could be extended to include different probability levels for different mutations types such as the probability of a BP sequence copy error being initated at that loci.
Just an idea with some proto-code behind it at this stage. No plans for any near term release.
Some discussion on this topic is
here.