It is understood that finding a "universal" functional is an NP complete problem. However progressively better functionals have been constructed decade after decade, and fitted to more and more comprehensive datasets. 

What is the closest humans have achieved, to a universal functional?