Another way of looking at this is that (unless there is a complete disaster) as you get closer to the end the uncertainty in the estimate decreases. When you get to the end, you know precisely how long it took.