We should save the raw data output to a file that is saved to repo when we update the recorded benchmarks, for more granular comparison.