Data at Cloudflare scale- some insights on measurement for 1,111 interns

This post details the challenges and insights gained from using Cloudflare's large-scale passive network data to detect internet censorship and middlebox interference. It highlights the difficulties of scale, noisy data (e.g., time resolution, DoS traffic, Happy Eyeballs), and lack of ground truth in passive observation. The author discusses the construction of IPTABLES rules for sampling TCP connections and the development of tampering signatures, emphasizing the limitations of passive observation in identifying the source of tampering and the scope of affected domains.