Mutations appear broken

[Testlund]

I've done some testing starting from version 2.44.1 and going back through the versions to try and figure out where the mutations got broken, preventing bots to evolve so they just sit on the screen doing nothing.
I started each sim with 32 autotrophs and 32 heterotrophs and ran it for 100 M cycles. Got the same results in all versions from 2.44.1 to 2.43.1L so far.

The only things that happens in the sims are that bots might shoot a little while, swim around, replicate a couple of times, form a tie to another bot, then nothing. No growth of the populations, instead they are either slowly declining or stay constant. It doesn't matter if costs are used or not.
I've checked Eric's explanations of the fixes between versions 2.43.1e to 2.43.1L to see where he might have done something that could have broken the mutations. It's pretty much only the inc or dec commands that rules the sims as far as I can comprehend.
Maybe someone here can explain it better what the problem might be. I can only go on what I see in bot behavior.
If someone could find the time to fix this problem I would be mighty grateful.  
Tonight I'm going to test version 2.43.1e.

[Numsgil]

I'm a wee bit conused.  By "broken", do you mean that bots' DNA don't mutate, or that they don't mutate to do the things that they used to?

[Testlund]

I mean the latter, in that the instructions seems to have been limited and the results are the same no matter what of the above versions I run. It was a long time since I saw bots filling up the screen or making lots of ties for instance.
I have to skip the testing of version 2.43.1e that seems to have a freeze bug.

[Numsgil]

Are you starting a fresh zerobot sim with each version, or are you trying to load up a saved sim you had from a previous version?

[Testlund]

I'm starting a fresh new zerobot sim for each version with the same settings.

[Numsgil]

I don't necessarily remember anything Eric did which might have done that.  Which version did you use last which definitely worked the way you wanted?

[Testlund]

I don't remember, but I'm going through the versions to see where it changed. Maybe it's not necessarily mutations that got broken. It could be some functions doesn't work. Maybe I can figure it out. This will take some time.

[Prsn828]

The answer:

The command indexing breaks any chances for useful values to evolve regularly.

The (possible) solution:

Greatly limit the chance of DNA values switching type, so that high indices are not lost.

The other (better) solution:

Change the source so indexing is not used to determine the meaning of values in DNA.

[Numsgil]

But that's always been the case.  It wouldn't explain any recent change in behavior.

[Testlund]

Maybe it could be a combination of what Prsn is saying and the fact that DB is not completely random. When more sysvars were put into the sim the program would change it's choices of what sysvars to pick when doing mutations. The various bool sysvars are pretty common in the dna for instance. I'm just guessing here.
Right now I'm testing 2.43o which looks a bit more promising. I got two bots shooting feeding shots while swimming backwards and one bot that emptied all it's energy in a snow storm. Things I haven't seen for a long time.

[Prsn828]

Certainly, I think the core of the problem is that mutations were simply a poorly designed feature in DB2.

DB3 will DEFINITELY fix this issue, and I expect evo-sims to be much more interesting.

[Testlund]

I can't wait!  
Another thing I've wondered about is the settings where you can move the slider between type and value of point and copy mutations and the various deviations. I'm not certain what effects these would have on mutations. I suppose type means various sysvars while value means the values stored in memory locations. The deviations I don't have a clue about.

[Numsgil]

type refers to things like the difference between numbers and add, div, mult, etc.  Different sysvars have different types.  So changing that slider means you change whether an add command is more likely to change into a mult or 42.

It was a reasonable idea at the time, but I think I probably over-engineered the mutations with tweakables.

[Prsn828]

In DB3 it is much simpler.

You get four text boxes with the chances of a certain mutation occuring placed in each box.
Any number (except imaginary) should work.

DB3 sums them up, and the chance of the result being type x is x/sum.
x can represent: command(s), constant(s), reference(s), store(s)

basically, you could say you want only constants to be added by point mutation, and only other types to result from copy error, and you get a bot whos DNA should only increase if those are your only mutations.

You also get a few boxes to input change length and probability.