Agree that a quality-sensitive metric can't be modelled fully in the structure. This is something I'm even observing in the work-environment also.

Do you have any learnings to make such initiatives work?