Design experiment for Group Calls with interference

Your team is adding a Group Calls feature to a 1:1 calling app. Design a robust experiment to measure its impact under likely network interference. Address: (a) Select exactly one primary metric and 3–5 guardrails; explain when "total call volume" is misleading (e.g., cannibalization, unequal exposure) and when it could be acceptable; propose user-normalized or rate-based alternatives. (b) Choose a randomization unit (e.g., graph clusters, ego-clusters, geo switchbacks) and a rollout plan that mitigates interference; compare bias, variance, and engineering complexity trade-offs. (c) If cluster-based randomization is not feasible, propose a quasi-experimental plan: e.g., invitation-gated treatment, time-based/switchback assignment, exposure-weighted estimators, IV using staggered access, or difference-in-differences with pre-periods; list assumptions and diagnostics you will run (placebo tests, balance checks, spillover detection). (d) Define exposure, triggers, and exclusion criteria (e.g., creators vs joiners; first exposure vs ever-exposed); detail contamination controls (e.g., invite tokens only visible to treated egos). (e) Pre-register analysis, specify minimal detectable effect with plausible baselines, seasonality controls, and how you'll handle novelty and ramp-up effects.