Power
Equal n per arm to detect a difference between two independent proportions with two-sided α and target power (normal approximation, unpooled variance in denominator).
Ignores clustering, multiple testing, and sequential peeking. For serious trials use a statistician and possibly simulation-based sizing.
