Define the hypothesis, primary metric, sample size, randomization unit, guardrail metrics, and decision rule before launch.

Good A/B tests avoid peeking, metric switching, sample-ratio mismatch, and interference between groups. The design should match the product decision.