Creatures Caves

That sounds very interesting... I've been fiddling a little bit with the brains in my development of CFF creatures, and I have learned a lot (at least, enough to make a lobe and tract that perform a function I wanted to add relatively well.. that is, getting the creatures to recognize and have a feeling of being full), but there is still a lot that is a mystery to me... specifically how the learning and dendrite migration/cementing works in their brains with the new state rules and just how reward/punishment affects these...

As you noted, a lot of state rules are undocumented or not very well documented... From what I can tell, the SV rule programming is register-based, similar to assembly language, with an accumulator for performing math functions, each neuron containing 8 registers, only a couple of which I have ever used or really know what they do... the accumulator seems to be able to hold fairly large decimal values, but the registers I am pretty sure can only hold decimal values ranging from -1 to 1 or 0 to 1... (which probably correspond only to a byte of information). I know that 'susceptibility' and 'Neural growth factor' in a neuron have something to do with dendrite formation and migration, 'state' is basically how stimulated the neuron is, 'output' and 'input' are registers that generally deal with values that are coming in from or going out to the dendrites in the tracts connected to that neuron...

Dendrites have an input neuron, an output neuron, and also seem to hold 8 registers, the only ones I really know what they do are signal strength, a value that tells how much the dendrite is being stimulated, and the weight, which determines how likely it is to migrate its 'end' to another neuron. I am not quite sure how dendrites that can migrate decide where and when to migrate though... I'm pretty sure it has to do with state rules in the tract as well as on the neurons in the lobes they are connected between. I know the state rules in the tract determine how the neuron at the end of the dendrite is manipulated based on the values in the neuron the dendrite is coming from.

An example of some relatively simple brain workings: In the stim lobe, there are neurons corresponding to each type of object the creature can recognize... It has tracts coming in from the various sensory lobes (smell, vision, noun)... These tracts define how dendrites are connected from the neurons in those sensory lobes to the neurons in the stim lobe. So, the stim neuron for 'food' has dendrites coming in from the sensory lobes, and the strengths of the signals connecting to it from those sensory lobes would affect the level of input the 'food' neuron gets. For example, if some food is closer, or smelling more, or moving around more, or the Norn just heard the word 'food', the input signal to the 'food' stim neuron would be stronger... This eventually will translate to the 'State' register of the 'food' neuron being higher, and that, along with all the other input to the stim lobe, will eventually help feed into the Norn's decision on what to pay attention to.

I have figured out a couple of state rules that aren't listed in your state rule list, though they were pretty obvious... load abs of, store abs in... just perform absolute value functions (the first takes the absolute value of a register and loads it into the accumulator, the second takes the absolute value of the accumulator and stores it in a register), bound in [0,1], bound in [-1,1] are math functions that just bound the value in the accumulator to those ranges... good if you are doing some kind of function that you want to have a cutoff point for, or if you are going to be storing your result in a neuron register or a dendrite's connection strength and want to make sure you are not out of bounds.

For example, I use "bound in [0,1]" in my new 'fullness lobe' in the CFF genome... I have a 'total hunger minimum' value I chose as a cutoff level for determining when the creature should start to be feeling full... then I subtract each hunger drive from it, and then that bounding function lets me make any negative result into 0 before I store it into the state of the fullness lobe's single 'how full am i' neuron (if the result was negative, that means the total hunger was more than my cutoff point, so the creature is not feeling full at all and the state of the fullness neuron should be 0)...

I then have a tract that goes to the creature's decision lobe, which connects dendrites from the single neuron in the fullness lobe to each of the neurons in the decision lobe that happen to correspond to a decision to eat something. This tract has an update rule that suppresses the decision neurons' 'inputs' based on the dendrites' 'signal strengths', which are set to be equal to the 'State' value in the fullness lobe... This makes whatever other factors that were making the Norn want to eat, that much less significant to it... So, if the creature is half full (hunger halfway between 0 and the threshold of no longer being full, so that the fullness lobe 'state' is at 0.5), all other inputs to any 'eat' decisions will be suppressed by half... If the Norn is totally full (all hunger drives at exactly 0, so that the fullness lobe's 'state' is at 1), then all the inputs to the 'eat' decision neurons will be totally negated... The creature couldn't possibly eat another bite and wouldn't think of eating.

You might also look up any information you can on the 'Curiosity Norns', and have a look at their genomes and the additional elements added to their brains (they have a whole new drive, a few expanded lobes, and I believe even an additional lobe and tract).