I also wonder if caching has some sort of influence. Maybe the spikes in the data are when the score is updated after a while of remaining static. The variation in this could have something to do with load levels on the site.