Little math problem(s) January 2019

[Allen]

In the case where the model was both slow and not precise, I would discard the model and do something else.

In the second case, where the errors are completely deterministic (but not precise), would it be safe to assume the underlying function is differentiable, and rather than look for variations in the output, one could change the input over successive trials, and use the information learned from that to refine the numeric estimate?

I can't imagine a real world modeling case where either of these two cases would come up. I could be misunderstanding the question?