I believe that one of the core tenets of multivariate testing is to capture as much data as possible.
As a product owner who's constantly testing, I often track a variety of goals with each experiment. Many of the goals are based off of custom events and sometimes just plain click goals. Another thing I typically also add is the goal called "Engagement: The percentage of visitors who clicked on any part of the experiment page."
These are all great, but sometimes I'm testing a button on a page that has many other things on it (and could also be dynamic and personalized). What I want to know is what percentage of users clicked on something OTHER THAN the button I am optimizing, for each variation. That would provide me with extra validation to determine if a variation is succeeding or failing.
I could set up events for every other link or button on the page, but again, there can be quite a few and it could be dynamically generated and personalized. The built-in "Engagement" goal also does not work because that will include the people that clicked on the button I am trying to optimize.
It could be set up on a per goal basis (where for every goal, you can see the % of users who completed that goal, and then the % of users who did something else). Alternatively, it could be set up on a per experiment basis (where it would show the % of users who did anything besides any of the goals that were attached to the experiment). Or it could have both.
Having a so-called Anti-Goal would allow me to see how people are interacting with an experiment in ways that I did not expect or anticipate.