It's the shortest, sure, but the best? If I saw this in production code I would be kinda miffed because the return function is indecipherable.

This is, in my opinion, an inadequately constrained kata. It never explicitly states that the order of the face (eyes, nose, face) elements will always be the same, you just have to infer it from the test data.

This kata is obviously written by someone whose's not well-versed in English. Skipping this one.

I definitely wouldn't qualify this as "best practices" since it's creating an entirely new string n times (via std::accumulate). This is an O(k * n) solution, when it should be O(n). For high values of n and k this is going to be sluggish.