We expect estimates to be spread around the target; i.e. the real results form a bell curve around our estimates. If estimates are consistently too short or too long, then we are not learning from our estimates.