Ad layout tests need a rollback path

Layout changes after an RPM drop should be narrow, reversible, and measured against the same segment.

Ad layout changes after an RPM drop should not be broad guesses. A safer test has a dated hypothesis, affected page set, rollback path, and baseline metrics. Otherwise the layout change becomes another variable and makes the next revenue chart harder to interpret. This matters especially for small publishers where one template or traffic source can move the whole average.