Ask HN: How do you handle unrelated actions in A/B tests?

Say we are A/B testing signups on our site, but some number of users come to our site, then login. Since these users appear to be irrelevant to our test, should they be removed from the final test results? Or does it not matter?

I can see arguments for either answer, but I was hoping someone on HN with a little more experience in the area could provide a stronger argument or even a statistical basis for one answer or the other.