Heh, the HR at my old company sent me too many bullshit candidates. We had already tried a solution where you'd solve multiple problems and then get assigned a score, but it sucked, every single one of them was an algorithm question.
I hacked out a quick page (that'd inform you it'll log everything, but gave you the problem to solve only after you accepted), that litterally asked you to solve fizzbuzz in any language you like.
But it logged timestamped keystrokes to the server and if the tab lost focus.
Of course I wrote a small player that would replay the logs so you could watch the candidate come up with the solution and watch them writing it :)
I just had too many candidates googling questions in a phone interview, so we changed our process.
They'd take the online fizzbuzz test, and then we'll invite them for an onsite interview if they pass, in my opinion it worked surprisingly well.
Too bad I can't try out how these guys are doing it.
With that approach, anyone who immediately switches to a programmer's editor to write the code and pastes it in when done (either manually or via something like It's All Text) will look identical to a candidate who searched for the answer and pasted it in. The former is a property you'd want to select for, while the latter is one you'd want to select against.
Yes. That's why i had a big fat warning telling them to not to this before they started the process :)
The page before they started basically told them everything we'd do. My line of thought was that anyone being able to bypass it is worth interviewing anyways.
I explicitiy told them that we'd log EVERYTHING and asked them to directly write the browser in the JS IDE (well, syntax highlight but no autocomplete) that'll come up, and that they shouldn't switch tabs or windows.
I still think this is fair, for litterally fizzbuzz, nothing similar or anything, just fizzbuzz.
I didn't care about (small) syntax/logic errors or anything (because you'd never run the code), but you wouldn't believe the amount of people unaware of the modulo operator.
Ah. If you're upfront about it, that's somewhat more reasonable. Though I'd still rather see more approaches that are tools-agnostic, and better approximations of real work.
Fair enough, I slightly amended my post, but I think asking for fizzbuzz with generic syntax highlighting is fair enough. As I think it's simple enough to write tool agnostic, because I'd expect you to be able to write it in any language you like, ignoring fatal syntax errors.
It's really just fizzbuzz, not an approximation of real life world, we had on site interviews for that (Switzerland is quite small, not sure how that'd apply to the US).
I hacked out a quick page (that'd inform you it'll log everything, but gave you the problem to solve only after you accepted), that litterally asked you to solve fizzbuzz in any language you like. But it logged timestamped keystrokes to the server and if the tab lost focus.
Of course I wrote a small player that would replay the logs so you could watch the candidate come up with the solution and watch them writing it :)
I just had too many candidates googling questions in a phone interview, so we changed our process.
They'd take the online fizzbuzz test, and then we'll invite them for an onsite interview if they pass, in my opinion it worked surprisingly well.
Too bad I can't try out how these guys are doing it.