The trouble is that this doesn't actually work. By using R this way you're assuming a model where every person is equally likely to spread the infection to every other person in the population, which is easy to calculate but what actually happens. In reality people can only spread the virus to those they're in contact with, and (say) a supermarket employee can spread the virus to more people than someone working from home and not socializing due to social distancing rules. Not only that, there are clusters of people who are all likely to spread the virus to more people or to less people when infected, such as workers in food manufacturing, people in dense urban areas vs less dense areas, etc...

So what actually happens in reality is that if R is hovering around 1, there are going to be some places where it's actually above 1 and cases are growing exponentially, and some where it's below 1 and they're shrinking exponentially. The end result of this is that places where R is actually below 1 make up an exponentially shrinking proportion of all cases, and as this happens it causes the overall measurement of R to go back above 1.