Think about better mapping between metric names in yaml and functions

(Premature optimization ...) Might not be worth it really.

But sometimes it currently does duplicate work, since we only run that on a small number of samples, it's not really noticeable. But I am sure it could be done smarter. I added some info in Readme