Design a system that continuously identifies the “Top K Hot Files” in Figma — the K most frequently opened or viewed design files in the last hour, day, and 7 days. The system must ingest a fire-hose of file-access events (tens of millions per second at peak), keep per-file counters that decay with time, and serve queries like GET /v1/files/top-k?window={1h|24h|7d}&k≤1000 with ≤100 ms p99 latency and 99.9 % availability. Exact counts are not required; a 1 % relative error is acceptable. You should support global rankings and be able to scale to 10× current traffic without redesign. Walk through your data model, streaming pipeline, aggregation logic, storage/caching layers, and the trade-offs you make between accuracy, freshness, cost, and latency.