I like this solution the most. Some questions on @wacnt (like this one) can't be answered because Wolfram Alpha's NLP fails to recognize the algorithm implicit in the question, but if it was able to translate the problem statement into a machine-readable form, this would be very close to it.

Edit: Some background on @wacnt: https://docs.google.com/document/d/17me2A-SG5Z0Mr9Xd-a1kT1HQ...