This is a typically poor example of ignoring the complexities inherent in analysing new user acquisition.

Suppose you implement a change that causes the registration form to crash for some subset of people? I am sorry to say for a while I was responsible for causing such issues at a very large service ;) Depending where the crash occurs and where you are counting signups you may actually get a massive boost in your metrics, as the victims retry with every attempt met by failure.

Suppose you correct an issue making it easier for people to login? Then lower registrations - the key here is to make sure you have a good metric to look at (e.g. accounts created that are interacting well with the site after as long a window as possible).