State-oriented consistency: Why we stopped looking for one right answer

A distributed systems team building a clustered message broker discovered that asking "which consistency guarantee does each piece of state need?" yields better designs than asking "which single consistency model should the whole system use?" The insight emerged after investigating memory exhaustion in their cluster—they found pods were loading entire session state datasets at startup regardless of which clients would actually reconnect to them, revealing a fundamental design mistake of optimizing for "the cluster can serve anyone" rather than "this specific state has this specific requirement."