How would you design an A/B test for a new recommendation ranking model at Handshake, including the unit of randomization and key pitfalls to avoid?
