A randomizer that biases itself to preserve uniformity of its outcomes without being predictably non-random. It's particularly useful for improving random placements without a large change in architecture such as implementing power-of-k random choices.