Rate Limits Protect State Integrity, Not Just Infrastructure

Rate limits are usually justified as capacity protection, but their quieter job is keeping write conflict probability low enough that state stays coherent.

Infrastructure survives bursts that shared state does not. The failure mode is not a crashed server; it is a workflow whose state stopped being explainable.

Sizing limits for integrity

• Cap concurrent writers per key range, not just requests per second.
• Watch conflict and retry rates as first-class limit signals.
• Give recovery traffic a reserved lane so incidents can drain.

Use the architecture guide, blackboard schema, and benchmarks to size limits around state coherence, not just server load.