I logged what actually happened here, which is why I am comfortable saying this. If I had to pick one factor, the ratio of active to dormant profiles accounts for most of the variance I saw.