The first attempt is a cold read and does not count toward the rating. From attempt two onward, average surprise is pulled toward the population average while the sample is small. Every new game replaces the previous rating, so more attempts help only when the result holds up. Handles are submitted by players and are not verified.