# Design Trending Collaborative Files Design a service that ranks currently trending collaborative files. Use the following scale and product rules as practice assumptions rather than claims about a particular product. ### Constraints & Assumptions - Events include qualified opens, edits, comments, shares, and unique active users. - Rankings are personalized only by access control and locale; the core trend score is global or workspace-local. - The list refreshes within five minutes and supports hourly, daily, and weekly windows. - Billions of raw events arrive per day. - Private file activity must never reveal a file to an unauthorized user. ### Clarifying Questions to Ask - What behavior should "trending" reward instead of merely "popular"? - Is the ranking global, per workspace, per locale, or per user? - How fresh and stable should the list be? - How should bots, repeated refreshes, deleted files, and newly created files be treated? ### Part 1: Metric and Contract Define a trend score, windows, eligibility, tie-breaking, and the read API. Explain how you would validate the metric before launch. #### Hints - Separate recent acceleration from lifetime popularity. #### What This Part Should Cover - Measurable product objective - Deterministic ranking semantics - Privacy and eligibility ### Part 2: Streaming Aggregation and Ranking Design ingestion, deduplication, windowed aggregation, top-k computation, storage, and backfill. #### Hints - Exact global sorting of every file on every event is unlikely to be necessary. #### What This Part Should Cover - Scalable event and window model - Candidate reduction - Late-event and replay strategy ### Part 3: Abuse, Reliability, and Serving Handle event spikes, hot files, manipulation, permission changes, stale data, and observability. #### Hints - A ranking can be available while a particular result is no longer eligible. #### What This Part Should Cover - Defense in depth against gaming - Fail-safe serving behavior - Quality and pipeline monitoring ### What a Strong Answer Covers - A trend definition tied to a product goal - Windowed streaming architecture with deterministic replay - Efficient top-k computation and permission-safe serving - Abuse controls, backfill, degradation, and online/offline validation ### Follow-up Questions - How would you personalize trends without creating a per-user stream job? - How would you explain why a file is trending? - How would you prevent a single huge workspace from dominating a global list? - How would you change the design for second-level freshness?