[SandwichTeeth]

I think that coding tests and challenges apply a sort of artificial pressure and requirement to have things memorized that aren't there in the real world. I think it would be a mistake to discount a candidate who has good code on their github when they had time to sit, think through problems without being under a magnifying glass, but trips up during a technical interview challenge/puzzle/whiteboard or whatever. Maybe in those candidates you could actually walk through their projects with them and try to determine if they have a deep understanding of what they wrote. To me, that seems a better indicator of how they would perform on a team with real problems, rather than having to puke up a pseudocode merge-sort from memory.