It's just not always true though, like in the case of "proper scoring rules". So it would seem at least theoretically possible to create a recommendation algori

It's just not always true though, like in the case of "proper scoring rules". So it would seem at least theoretically possible to create a recommendation algorithm such that trying to "gamify it" would only increase the utility of its output.