Now, logically, as these super frogs reproduce etc, the population becomes specialised to area. Some frogs only survive if they are large, others if they are small, others again if they are green, others if they swim fast. Whatever the reason, different features of the population A are 'bred out' due to being unfit.
Draw these four circles within the A circle - again extend by drawing dotted lines around them.
It is completely normal, and indeed expected, that the area covered by A and A' significantly outweighs the area covered by the small specialised populations. The combinatory possibilities have been reduced. Multiply this by a large number of generations and you have an instant problem. Diversification is supposedly increased - you have a large number of different types of frogs - but in actual fact genetic potential has been lost.
Thank you for taking the trouble to spell out your argument.
You have overlooked several factors in your approach. One fundamental mistake is that you are using a model to draw conclusions about how the natural world operates, without confirming that you have included the essential features that you are interested in. (As statisticians say, all models are wrong, but some models are useful.)
The first factor that you have overlooked is that your loss of genetic diversity does not correspond to any real biological situation, at least not one that has occurred at any point in the last several billion years. You never start with a population with a wide range of frog genotypes unrelated to fitness -- where would you get such creatures?. Real populations are almost always already well adapted to their environment and lifestyle, so there is no big spread of suboptimal genetic diversity to lose. You
start with all frogs already clustered around local fitness peaks; that's the equilibrium situation, so the loss of diversity you're talking about never occurs.
Second, your model assumes that there are simple local optima, that all frogs in a region of genotype space will converge on a single optimum genotype. Real fitness peaks aren't like that: there is always variation in fitness on a fine scale, both in space and in time. Frogs that are ideally suited to this year's pond will have offspring that are less well suited to next year's pond, or to the part of the pond they happen to end up in. This means that real species are clouds in genotype space, loosely clustered around broad peaks. The varied and varying environment ensures that a range of genetic variation will continue to be present in the population. Plus, of course, new mutations are constantly pushing the cloud outward, even as selection prevents it from expanding too far.
Third, you have neglected the effect of frequency-dependent selection, and more broadly the dependence of the fitness landscape on the genotypes. Organisms generally have to compete for resources, both with conspecifics and with other species; a genotype that was most fit for the first 25 frogs in the pond may be less fit for the 26th, which might be better off preferring a different food source, for example.
What this means is that, in the real world, there is pressure not only for organisms to evolve toward the local optimum, but also to evolve away from the local optimum (optimum only in the absence of competitors) in order to fill a new niche. So there is a tendency both toward loss of diversity ("purifying selection") and toward speciation and radiative expansions.
This IS happening with every population round the world. We are getting greater "specialisation", but at the cost of lower information potential. Kinda like kinetic versus potential energy.
I'm sorry, but there is zero empirical evidence that this is happening in any systematic way in the real world. If your model says that something must be happening, and it isn't happening, the problem lies in your model.
Why do you think mongrels have less disease problems than particular breeds? It is purely because the information loss is not the same. There is less likelihood that the information necessary to combat disease X has been bred out of them.
Dog breeds are the result of extremely intense selection, operating on small populations in the very recent past. This means that many deleterious alleles have been fixed in breeds, sometimes because they were directly favored by the breeders, and sometimes because they were linked to favored alleles. Nothing like the process has happened to wild dogs, has it? Have they been splintering into genetically impoverished specializations?