It seems like all "code exercise" sites like this have the same annoying problems which I'll generally describe as lack of attention to detail in the exercise descriptions.
Take their Fibonacci exercise as an example. In the first section they say, "The first line of the input will be an integer N (1 <= N <= 100)" which specifies how many test cases follow. Then in the next section they say, "The first line of the input will be an integer N (1 <= N <= 10000), specifying the number of test cases"
So besides the fact that they aren't even consistent in specifying the expected range of N it isn't even clear why you would need to specify the number of test cases in the first place vs just reading one from each line until EOF. That makes you think maybe they want you to put in some basic error checking which again since they aren't consistent with N turns into a trial and error exercise if it even matters at all.
There is also no indication about the version of interpreters or compilers being used to check the submissions. If I choose to write my solution in python is that python 2 or 3?
Combine all that with an implied scoring system based on speed or number of tries and you have to wonder if this is really measuring anything relevant or if it is just filtering for people who's default assumptions happen to be the same as the person who wrote the lame exercise description.
I had to seriously do a double-take, I thought this was straight trolling, not a real company.
I doubt any serious developers would consider doing this, there's already a way to demonstrate your programming capabilities in the real world, either by contributing to an existing open source project or maintaining your own projects on github.
I agree: I have no idea how the scoring system works. Am I supposed to get it done in as little (real world coding) time as possible? As few keystrokes? Is it supposed to be done in as few cycles (on their test machine) as possible? Should it be bulletproof, or just work?