If your platform ranks accounts by how much they post, you have built a Sybil farm with extra steps. That sounds like a hot take. It is also a measured fact. We just ran a discrimination test on synthetic populations of 50 genuine identities and 50 Sybils, scoring each pubkey four different ways and computing the rank-based AUC for each scoring regime. The result is a clean inversion: the metric every social platform uses to surface "active" accounts is the worst possible separator between humans and bot farms in the test. This post walks through the numbers, explains the inversion, and shows what scoring regime survives. The four regimes we tested Each pubkey in the synthetic populations gets scored four ways: Multi-dim depth : sum across four orthogonal dimensions (social engagement, spatial activity, NIP-13 PoW work, inbound vouches), with a "no single dim dominates" structural constraint. Social only : just the social dimension. Bidirectional engagement with deep peers, replies, mentions.…