Build the Instagram News Feed: a personalized, ranked, infinitely-scrollable list of posts from the people you follow, blended with recommended content. The product is read-dominated (≈ 60:1 read:write — feed-opens 350K QPS peak vs post-creates 5.6K QPS peak), personalized per user (no edge cache for ranked JSON), and ranking-bound (the model is the product — chronological feeds have 30% lower session length per Meta A/B disclosures). The architecture is a two-stage retrieval funnel (candidate gen + early-stage ranker → late-stage ranker), a hybrid fanout (denorm-on-write for normal users, pull-on-read for >100K-follower creators), and cursor pagination (no OFFSET — at depth 1000, offset is 17× slower per the standard cursor-pagination analysis).

This canonical is sized for ~1.5B DAU, derived from public Meta scale figures (Q4 2023 earnings; Mosseri 2021 IG MAU disclosure) and the published architectures it builds on: TAO (USENIX ATC '13), Memcache + leases (NSDI '13), Haystack/f4/Tectonic (OSDI '10/'14, FAST '21), the Instagram Explore recommender system (engineering.fb.com 2023), and Manhattan-style sharded counters (Twitter blog).