These graphs show the results of evolutionary experiments by Richard Lenski, in which a bacterial species ( e coli) has been evolving under constant conditions for many years: tens of thousands of generations.
These bacteria were not perfectly adapted to the experimental environment, and so there is selection for changes that allow them to do better in these conditions. Adaptive change is rapid at first and slows down with time, as the culture approaches an optimum phenotype. Fitness increases rather like the logarithm of time.
The probability of a beneficial mutation fixing is proportional to the advantage it confers. Large-effect beneficial mutations are more likely to fix and dominate the early phase. As the bacteria get closer to an optimum, the possible gain from a beneficial mutation is smaller, and so those smaller-effect beneficial mutations ( the only ones possible) are less likely to fix. Thus they take longer to fix (on average they need to occur many times before succeeding) and they also fix more slowly, since their growth advantage is small.
relevance: a new virus in humans is like the situation near the origin of graph B. The virus is not yet close to an optimum, so change is fairly rapid – particularly if the virus is infecting vast numbers of people ( like covid-19) which greatly increases the number of copies of the virus and thus the chance of favorable mutations ( Fisherian acceleration). Favorable to the virus, that is.
An old virus in humans, say measles ( > 1000 years old) is closer to an optimum: change is much slower.
It seems that most professional virologists are used to viruses that have been around for quite a while – understandable, since new viruses do not sweep through the human race every year.
You could have predicted the emergence of new higher-transmission variants of covid-19 from this theoretical perspective. I did, arguablywrong did, probably others have as well. But virologists did not.