The peak is in April of 2022. On the surface you could take this as a graph of the relative popularity of Clojure, but I think that would be way too reductive. I am interested though how to explain some of these trends and numbers. Is the clojars reporting accurate? How does tooling and caching come into play? After all downloads does not reflect how often these things run. New releases would increase downloads, etc.

Up to 2022 it's nice to see though, a nice story of steady growth, but what comes after I'm less certain about.

Maybe back in 2022 people finally figured out how to setup CI caches?:)

Interesting! These stats are derived from Fastly logs (before 2016 it was nginx access logs, since all artifacts were served from disk). Re: accuracy - I think the stats are fairly accurate for requests that actually hit the CDN, though there have been times where we have lost data due to lack of disk space, bugs, etc (I believe that is what happened with the big dip at 2021-02). I know there is at least one big mirror in Asia that reads directly from the s3 bucket, bypassing the CDN, so those stats wouldn't be reflected here.

I guess to further investigate it would be good to look at the package level. Is it specific packages that get downloaded less, or is it a general downturn?

if anyone wants to poke at it here's some code to load all the edn files into a postgres table https://github.com/plexus/clojars-stats/blob/master/stats_db/src/stats_db/core.clj

that repo also contains a copy of all the download edn files