Thinking about how to measure this properly, why not just the moving average of daily downloads over 30 days from each repository?

… yes CI would be a lot of these downloads, but it’s at least a useful proxy