How to invent an interesting problem which can be solved in 2 hours, which can not be found in the Internet and would allow candidate to demonstrate something?
And you need to produce such problems relatively often, right?
Ideally you only produce one good one, which allows you to benchmark answers against previous ones. The other important thing is that you should do the task as well - that both ensures it is actually doable in the allotted time, and gives you an initial benchmark of what you're hoping for.
And you need to produce such problems relatively often, right?