How would you design the frontend infrastructure for an A/B testing platform that ensures consistent variant assignment, avoids flickering, and tracks statistical significance?

Imagine you and your friend get different ice cream flavors to see which one you like more. A/B testing is like that: we show some people a blue button and others a red button, then count which one gets more clicks. We have to make sure everyone always sees the same color and that the switch happens super fast so no one sees a flicker.