Prove friends outperform unconnected; design experiments and metrics

Using info_stream_views and post_reactions, design a rigorous analysis and experiment to determine whether content from Friends is "more social" than content seen by Unconnected viewers, and to quantify the additional value of Unconnected exposure even if its engagement is lower. Part A — Define and justify metrics: - Precisely define "more social" using measurable metrics derivable from the tables, e.g., reaction rate per view, comment rate per 100 views, reshare rate per 100 views, dwell-time lift (e.g., share of views >60s), and conversion-to-reaction within same day. - Specify unit-of-analysis (viewer-post-day vs. post-day) and whether to include zero-reaction views; explain how your choice affects bias. - Propose normalization (per impression, per unique viewer, per minute viewed) and guardrail metrics (e.g., daily active viewers, view diversity across creators). Part B — Causal inference plan (observational): - List key confounders (viewer propensity to engage, author popularity, post age, time-of-day, device) and describe stratification/matching or inverse-propensity weighting to compare Friend vs Unconnected views holding these constant. - Describe how you’d cluster standard errors (by viewer and/or post) and handle repeated measures and multiple comparisons. Part C — Experiment design: - Propose an A/B test on ranking weights for relationship (e.g., downweight/weight Friend vs Unconnected slots) with randomization at the session or viewer level. Define primary outcome(s) and guardrails; specify sample size assumptions (baseline rates, MDE) and test duration; include a pre-registration plan and power analysis approach. - Address interference/spillover (same post appearing in both feeds), novelty effects, and saturation. Propose a design variant (e.g., post-level randomization or two-sided marketplace experiment) to reduce network interference. Part D — Value of Unconnected beyond engagement: - Define and measure discovery value (e.g., new follows to creators, breadth of creators viewed, topical diversity/entropy), long-term retention or session depth, and reshare-driven reach. Propose proxy measurements using only the given tables and call out what additional logs you would request. Deliverables: a) a metric spec with formulas; b) an observational analysis plan detailing controls and diagnostics; c) an experiment design doc with randomization unit, power inputs, and stopping rules; d) a set of KPIs quantifying the incremental value of Unconnected content even if near-term engagement is lower.