We should add a way to run the benchmarks on a branch of a package, compare with the last updated `master` report, and post a Markdown summary.