I really like where Assaf has gone with this, and am interested to see what he comes up with as time goes on.
I'd love to see this evolve into something that uses hardware and software decoupling to enable more effective and deep experiments.
I'm curious about why adding two additional fields to the registration actually improved the conversion rate. Was this just an example, or does someone have some ideas as to why this happened?
It's just an example. The idea of A/B testing is to prove what works better, and it doesn't always follow conventional wisdom. This simplistic example is probably not true, but that's why it's a good example.