What about performance? E.g. you don't want to implement an O(n) algorithm if an O(log(n)) equivalent exists, and merely testing inputs and outputs won't reveal that.
I do agree, but it depends on the criticality of performance and the level of change the system goes through by default.
For us we are always implementing complex new logic for not very many users, so there's not much benefit to some clever data structure that we are all too dumb to understand. If I get a ticket that involves inserting something into the logic you wrote, and I can't understand how to operate your current solution, I will be hacking up an n logn solution as long as it is adequately performant
