Work out feature flag exposure instantly with clear inputs, formula shown and shareable results.
A feature flag serves two purposes that pull in opposite directions: limiting blast radius and measuring effect. The standard approximation for a two-proportion test needs about 16 p(1-p)/delta^2 users per arm, so detecting a 5 percent relative lift on a 4 percent baseline takes roughly 240,000 users — far more than a cautious 10 percent rollout provides.
Exposure and power
exposed = users x rollout; sample per arm ~ 16 p(1-p) / (p x relative MDE)^2
Rarely. Use a small rollout to verify it does not break, then raise it to a statistically meaningful share to measure the effect.
They must be removed. Long-lived flags multiply the number of code paths, and each combination is a configuration nobody has tested.