What is the app architecture for Facebook and Netflix that enable extreme multivariate testing? (No two users' UIs will look alike)

My initial thoughts: maybe categorize users into test groups, and then have a dedicated interface for each category? The problem with this breaks the DRY convention.

So maybe each user has a bunch of feature flags like

{
show.featureA:true,
show.featureB:false,
....
show.featureN:true
}

And then the server applies a filter based on those flags?

Hope this kind of makes sense. (I'm a garbage dev)