You're testing a feature in a collaborative product where one person's treatment can affect teammates through shared projects, task updates, and notifications. The team is debating whether to randomize users, projects, or whole workspaces.
How would you design an experiment for a feature with strong spillover effects across teammates?