You work on a product team that is considering a change to feature X, and the team wants to know whether it is actually affecting metric Y. You have enough traffic to run an online experiment, but the result needs to be credible enough to support a launch decision.
How would you prove whether the change in feature X is affecting metric Y? Walk through how you would define the test, choose the right metric and unit of randomization, and decide whether the result is strong enough to ship.
