On 09/10/2007, Gregory Maxwell gmaxwell@gmail.com wrote:
In addition to any large classes, I think it would be useful to also apply it to some number of pages quasi-randomly selected. Probably something around 10000 pages would make sense. We could set a study end date and undo the setting at that point and also gather data on page activity after the change is removed.
10000 isn't enough to prevent one determined person messing with the results.