The biggest issue I have is that you want the solver to guess what's in your head. In this case it is woefully inefficient as Fibonacci numbers can be generated in logarithmic time while both bottom up and top down are linear.

On a separate note, there are plenty of recursion and Fibonacci-related katas (e.g., this: https://www.codewars.com/kata/59928e2889d6a01970000051) that test for efficiency. There are also some bottom up vs top down katas; the one that comes to mind is https://www.codewars.com/kata/the-fusc-function-part-2