experiment-designlisted

# Experiment Design ## Goal Design product experiments that produce decision-ready evidence without confusing statistical movement, business impact, and launch approval. ## Workflow 1. Write the hypothesis in a testable format: audience, intervention, expected metric movement, mechanism, and minimum useful effect. 2. Choose exactly one primary decision metric. Add guardrails for quality, retention, revenue, support, privacy, reliability, or compliance risk. 3. Define population, eligibility, exclusions, randomization unit, exposure event, analysis window, and timezone. 4. Estimate feasibility from baseline rate, traffic, minimum detectable effect, expected duration, and instrumentation readiness. 5. Select the experiment type: A/B, multivariate, holdout, sequential rollout, fake-door, prototype test, concierge test, or beta cohort. 6. Define stopping rules before launch. Avoid declaring success from early spikes or unplanned segment fishing. 7. Add data-quality checks: sample-ratio mismatch, missing exposure events, bot/internal traffic, delayed ingestion, release overlap, and instrumentation drift. 8. State decision rules: ship, iterate, stop, extend, or investigate. 9. Keep launch-sensitive approvals separate from experiment success. ## Boundary Use this skill for experiment design, rollout tests, fake-door tests, beta cohorts, and experiment decision rules. Use `skills/product-ops-analysis/SKILL.md` for exploratory data analysis, `skills/metrics-tree/SKILL.md` for ge