Design an experiment with marketplace network effects

Uber is preparing to introduce a networked product in a two-sided marketplace that connects riders and drivers. How would you set up a causal experiment that handles interference or network effects? Specify: (a) the randomization unit (user, driver, city, or geographic cluster) and your reasoning; (b) how you will reduce and quantify spillovers (for example, cluster randomization, geographic lift tests, partial-interference assumptions, graph cuts or holdouts); (c) the primary and guardrail metrics and how you would calculate them; (d) power and sample-size calculations under clustering, including ICC and design effect, plus the expected run time; (e) bias-reduction methods such as CUPED or pre-period stratification, and how you would handle novelty or washout effects; (f) diagnostics you would run to identify SUTVA violations and noncompliance; and (g) how the design would change if driver supply were effectively unlimited. Give a concrete rollout plan and analysis outline.

Overview: This question tests a candidate's ability to design causal experiments for networked two-sided marketplaces, with emphasis on managing interference and spillovers, choosing cluster randomization, defining metrics, computing power and sample size, applying bias-reduction approaches, and running diagnostic checks.