If there was a standardized way to save this information, and tie it up to each commits, it would be insanely useful to amass a very valuable training dataset.