People often quote a single perf number when comparing versions (e.g. "Python 3.11 is 25% faster than Python 3.10")

The pyperformance compare command prints out the "change factor" per benchmark. It's not clear how to go from this list of numbers to a single overall number. Having the compare command compute and print it out would also help with consistency when quoting this single perf number (is it mean? geometric mean? only including significant results or all of them?)