Plan an A/B test before you run it

Help me write a one-page plan for an A/B test, so that {{team}} agrees on what success means before the test starts and nobody reinterprets the results afterward.

The change we want to test: {{change}}
Why we think it will work, and the evidence: {{rationale}}
Where it runs and who sees it: {{surface_and_audience}}
Current baseline for the main metric, and typical weekly traffic: {{baseline_and_traffic}}
Constraints, such as launch dates, other tests running, or legal and privacy limits: {{constraints}}

Write the plan with these headings:
1. Hypothesis: "If we [change], then [metric] will [direction] for [audience], because [reason]." One sentence.
2. Primary metric: exactly one, with its precise definition.
3. Guardrail metrics: 2 or 3 that must not get worse, each with the threshold that would stop the test.
4. Design: the variants, the unit of randomization (user, account or session), the traffic split, and who is excluded.
5. Size and duration: the smallest effect worth detecting and the inputs a sample size calculation needs. Give a rough duration only if the baseline and traffic above allow it, show the arithmetic, and tell me to confirm it with a sample size calculator, because rough estimates can be far off.
6. Decision rule: which result means ship, which means do not ship, and which means extend, written before any data comes in.
7. Risks: up to 3, such as novelty effects, seasonality, or a change that also affects users in the control group.

Keep it under 500 words, in plain language a non-analyst can follow. Do not choose a metric because it is likely to move; choose the one that reflects the goal. Avoid adding secondary metrics beyond the guardrails, since every extra metric raises the chance of a false win.

If the change cannot be tested this way, for example because traffic is too low or the change cannot be shown to only some users, say so and suggest a better way to evaluate it.