IRC logfiles, mostly. You're seeing examples of code snippets typed into IRC from the last 4-5 years run in a sandbox under Clojure 1.5.1. I also ran it over the ClojureDocs s-expressions. There are a lot of examples missing (defn, def, etc. for instance) due to the fact that I didn't want to inadvertently run something evil, but FWIW there are over 30k examples of sequence functions, JVM interop, etc. In a previous comment I mentioned that my plan is to add the ability to submit, edit, bookmark, and rate examples.
RE: GitHub projects tagged "Clojure": I think exploring Codeq (https://github.com/Datomic/codeq) would be a cool way to go about doing that kind of thing.
Last time I tried codeq, the analyzer was limited to symbols defined outside of functions. The Clojurescript parser might be a better option, because the parse tree would allow indexing usages inside functions.
I also wonder if a generative testing library could be used to generate examples.
What is it scraping? Is it just Github projects with language "Clojure"?