Instruction file imported from iPzard/coding-agents (.github/instructions/architecture-scalability/cap.instructions.md). Copyright stays with the author.

CAP

Cite as: [CAP]

Principle

During a network partition, you choose. Linearizability ("read sees latest committed write") is expensive and global. Most operations don't need it. Code that assumes linearizability across a distributed system (e.g. "we just wrote X, immediately read it back from any replica and expect X") is broken under partition.

Red flags in code

• Distributed read-after-write logic that assumes immediate visibility.
• Cross-region or cross-replica consistency assumed without explicit coordination.
• Code that breaks badly during a partition because it doesn't tolerate stale or unavailable data.

Apply directly (mechanical, unambiguous)

• None — CAP-level concerns require design adjustments.

Propose only (requires judgment)

• Choosing CP vs AP behavior for an operation.
• Designing a fallback for partition tolerance.