[peteretep]

I used to give people a description of how an algorithm was implemented, and ask them to write it out in code. Specifically we'd take the conversation to boundary conditions, failure conditions, testing, etc, from there. I was reasonably happy with the outcome.