I want to note that there is not scaling factor. What we did is to replicate all the computation using the Fedora methodology with our datasets. In other words, we do not convert the UUID way of counting: for the comparison we discard this result and redo all using IPs and in the way describer by Fedora's scripts.
This is the reason that we do not have errors or confidence intervals here. But could make sense for the linear regression model that we used to extrapolate the evolution of the distribution. But this model is not related with Fedora comparison.