##### Scenario A product team at Meta wants to understand whether content from a viewer's friends (connected authors) drives more 'social' engagement than content from unconnected authors. The team is about to introduce unconnected content into the Info Stream (feed) for the first time and needs to know how to validate the hypothesis and measure whether the launch succeeds. You have access to impression, view, and reaction data (reactions, comments, reshares), the friend/connection graph, and content metadata. ##### Question Work through the following parts: 1. **Validate the hypothesis.** Using the available views and reactions data, how would you validate that friends' content is more 'social' (drives stronger engagement) than unconnected content? Outline the metrics that define 'socialness', your hypotheses, and the statistical tests you would run. Detail the concrete steps to measure engagement for connected versus unconnected audiences, controlling for confounders such as feed position and content type. 2. **Design the launch experiment.** Unconnected content will be introduced to the Info Stream for the first time. Focusing on metrics and experimentation, explain the experimental design (randomization unit, arms, duration, power) you would use to measure whether this launch is successful. Specify primary success metrics and guardrails (e.g., retention, time spent, negative feedback, friend-to-friend interactions). 3. **Interpret the results and decide.** How would you interpret the experiment, including cannibalization/displacement of friend engagement and segmentation effects, and what would your launch decision and rollout plan be? 4. **Quantify the alternative value of unconnected audiences.** If per-impression engagement from unconnected viewers is lower, what alternative value could they (and unconnected content) provide, and how would you quantify it? ##### Hints Define socialness (e.g., reactions/comments/reshares per impression); use within-post or within-user comparisons with fixed effects / propensity methods to control for position and selection bias; run a user-level A/B test with dose-response insertion caps; track guardrails (retention, time spent, friend interactions, negative feedback); for alternative value consider reach, discovery, network growth, and creator health.