If you provide novice problem-solvers with a problem, they’ll attempt to solve it using superficial strategies, comparing it to routine problems that they already understand. This much I have covered already. But if you provide novice problem-solvers with — instead of a problem — a set of constraints, and then ask them to form and solve their own complex problems, something amazing happens — they solve these problems with expert-level strategies.
It’s the difference between teaching someone to fish, and having them invent things to do around a pond with a hook line and stick.
The strategy Gage described here for helping players learn games is as genius as it seems obvious in hind sight.
The big thing for me is that if you want to add this to your game it needs to be sufficiently complex to begin with for sandbox to be interesting. Which is a great litmus test for if your game is good in the first place.