Let's unpack this: v4 has a host of changes 1) new code 2) ranking algorithm 3) ui The new code and ranking algorithm are intimately tied. The key problem is

Let's unpack this: v4 has a host of changes 1) new code 2) ranking algorithm 3) ui The new code and ranking algorithm are intimately tied. The key problem is that any election-based website has a single "correct" view, making phased beta difficult, but not impossible. This is complicated by the fact that Digg is a startup, and doesn't have infinite amount of money to solve this. So, ideally, what is required is 1) a parallel run of the two algorithms, measuring some subjective quality of the results. (e.g. click throughs) 2) Notifying users that personalisation is being progressively turned on. Thus, breaking up the audience into cohorts (i.e. no single true view). 3) Using Eric Ries' style AB testing to continually tweak and split test, rather having a big deployment.