I sat with this question for a while here, and stopped assuming the obvious answer was right. The pattern was clear enough: how the ranking rewards activity accounts for most of the variance I saw. The evidence pointed the same way without exception whether the app was well known or not.