If they are that rare that they have not been observed, then I would start to feel like one of the core foundations for evolution is resting on quicksand -- inferred not observed.
Not really. Darwin, after all, developed his theory with no knowledge whatsoever of mutations. All he knew was that within any population, characteristics varied.
Today, we know that varying characteristics have a genetic basis. But we also know that there is no simple correlation of genes with characteristics. The simple high-school examples based on Mendel's experiments would lead one to believe there is one gene each for flower colour, seed colour, seed texture, etc. But in most cases a gene has multiple effects and each characteristic depends on the interaction of multiple genes. Furthermore how a gene will express can be further modified by regulatory factors and environmental conditions. This makes the correlation of genes to characteristics extremely complex and difficult to study.
Mutations, or changes in gene sequence, add another level of complexity. Sometimes a mutation has no effect at all on the protein product. Sometimes it may substitute one amino acid for another, yet change nothing in the functioning of the protein. Sometimes a small change, especially in a gene that regulates embryonic development may have a remarkable effect.
So basically, there are several levels of complexity leading to variations in characteristics and we are only at the threshold of disentangling them. How do mutations affect genes? How do genes affect protein products? How do regulatory genes affect other genes? How do any of these affect metabolic, physiological, morphological, developmental and/or behavioural characteristics.
It is these complexities that make it difficult to tie the observation of a specific mutation to a specific change in a characteristic. It is much easier to observe the variations themselves and see how they are affected by natural selection. But sometimes we can see a variation and not be able to identify the genetic difference(s) that generate it, and sometimes we can see genetic differences without being able to determine what difference, if any, it makes in its possessor.
Nevertheless, we also have sufficient knowledge of genetics to know that the variations are the consequence of genetic differences. And to know that mutations do affect genes and how they operate.
The whole matter gets another level of complexity from the fact that there is a lot of motivation to discover a treatment in the case of things going wrong, and not a concomitant commitment to researching why things go right. So the bulk of investigation into the effects of mutations has been into those that cause problems, whether in ourselves or in the domesticated animals and plants we depend on. Hence we know a lot more about mutations that cause problems than about those that do not.
Finally, we should note that really there is no such thing as a good or bad mutation. Mutations are not good or bad in themselves. It is the consequence for survival and reproduction that is important, and that is not directly determined by the mutation per se, but by the impact of the environmental conditions. "Good" or "bad" always has to be understood in terms of this relationship. And that actually brings us back to variation--the observable differences among individuals in a population. Just where Darwin began.
Variation is what natural selection acts on directly. It is at the level of variation that the interaction between creature and environment takes place. Indirectly this has implications for genes and their mutations, since they are the source of variation. But connecting the levels between mutations and variations is not easy. It is a huge learning curve requiring a lot of research, and we have only just begun.
