CP scenario (Consistency + Partition)

A financial ledger system that processes inter-bank transfers requires strict correctness even if the network splits. During a partition, the system must reject writes rather than risk divergent account balances. Reads must always return the most recent committed state from a majority quorum. After the partition heals, no reconciliation steps requiring human intervention should ever be needed because the system never allowed conflicting writes.

AP scenario (Availability + Partition)

A global social network feed service must keep accepting posts, likes, and profile updates even if regions lose connectivity. During a partition, all nodes must allow local writes using last-write-wins timestamps. Clients must always get a response, preferably from the nearest replica, even if it returns slightly stale data. After the partition heals, automatic conflict resolution must reconcile divergent versions without blocking front-end traffic.

EL scenario (Else: Latency over Consistency)

A real-time product recommendation engine must return suggestions within 50–70 ms for every user request. Under normal operations (i.e., no partition), the system must prioritize returning locally cached or regionally stored results without waiting for synchronous cross-region replication. Slightly stale recommendation models are acceptable during normal operations as long as latency stays within SLA.

EC scenario (Else: Consistency over Latency)

A fraud detection rules database must always return the most correct and up-to-date ruleset during normal operation. Even when the system is healthy, any update to rules must be synchronously replicated across regions before a read is served. This increases read latency, but correctness must be guaranteed at all times unless there is a network partition, in which case write-blocking is acceptable.

CA scenario (Consistency + Availability but only possible without partitions)

A single-region relational database running inside a highly reliable isolated datacenter must guarantee both correctness and full availability because the environment assumes partitions do not occur. Requirements include ACID transactions, zero write-loss, and immediate read-after-write correctness. Failover to replicas inside the same LAN must happen within milliseconds without violating linearizability.

PA scenario (Partition + Availability but inconsistent data allowed)

A distributed caching layer for a ride-hailing app must continue serving location and ETA updates during regional outages. When a partition happens, nodes should prefer local writes and local reads to maintain responsiveness. The system accepts that different regions may show conflicting driver locations temporarily, with a background reconciliation once connectivity returns.

CL scenario (Consistency + Low latency not achievable; therefore consistency with slower responses)

A configuration management system for critical infrastructure may prioritize correctness but has no strict latency requirement. Clients must be prepared for read requests to take long due to synchronous cross-datacenter coordination. The requirement is to ensure absolutely no stale configuration values reach machinery, even if requests become slow during heavy load.