Common misconceptions about randomness underlie the design and implementation of randomness sources in popular operating systems. We debunk these fallacies with a survey of the "realities of randomness" and derive a number of new architectural principles for OS randomness subsystems.