In this background I will explain two things: how genetic algorithms (GAs) work and how polyga is implemented. I feel you should understand some of the basics behind GAs first so you understand why people use them, and then I can explain how polyga is implemented and why it is implemented in this way.
If you already know a lot about GAs and only care about how polyga is implemented, see polyga.
All genetic algorithms work based off the circle of life and Darwinian evolution.
At the top of the circle is selection. This is where certain animals (bears in this case) are selected to breed based on environmental (or cultural) factors. In this case, global warming (an environmental factor) is causing temperatures to rise, killing our poor polar bears that are maladapted to warm environments. Thus, only brown bears breed during the next "crossover" phase.
In this phase, animals (or polymers) breed, exchanging genetic material and creating new children, potentially more suited to the new environment. In nature, this crossover is of genes, but for us it will depend on the design task. For instance, GAs can be used to tune hyperparameters of machine learning models, in which case the values of those hyperparameters are "crossed over" or mixed. For polymers, chemically unique blocks are crossed over (seen in the image below). The main point is, this crossing over generates a new child that may have the best (or worst) features of both parents.
After crossover, mutation occurs. In this phase, some genes in an animal are mutated, causing potentially beneficial adaptions to occur. In the case of our brown bears, they now grow antlers, which admitedly might not be very useful, but they sure are easy to make cartoons of!
Finally, each child is ranked according to some fitness function. In real life, this fitness function is a complex function of the predators in an environment, terrain, disease, culture and more. Obviously this would require a lot of computing power and code to replicate, so our GA is likely going to be a much simpler.
This was a very basic background, if you're still curious about GAs there is a lot more you can read, but I suggest trying to use a GA first.
Once a basic GA finds a space that achieves the target, it tends to get stuck. It doesn't explore a wide space anymore, staying in the local optimum.
I've found that if you don't add some restrictions or way to assess realism, they can make some wild polymers. I'm sure this is the case for other design problems as well.
Selection techniques include things like elitest selection, stochastic sampling, roulette wheel, tournament selection and more. Crossover schemes include single point crossover, double point, k-point, uniform, and more. Mutation schemes include displacement, simple inversion, scramble mutation and more. My point is, polyga implements some of these and not others, as will most GAs you find out there. These will influence results, so I would keep in mind how any algorithm you use works if it doesn't seem to be effective. These might influence results and you could try implementing a different scheme if you're not satisfied.
With those caveats stated, let's describe how polyga is implemented.