When to Split, and When Not To

1. The Triggers That Are Real

A split should be caused by something observable. Five conditions genuinely justify it, and each can be checked rather than argued about.

• Deploy coupling is limiting delivery. Teams are ready to ship and cannot, because one deployable carries everybody's changes and a revert reverts work that was fine. The measurement is whether deploy frequency is limited by coordination rather than by readiness.
• Scaling needs have diverged. One component needs far more resources than the rest, or different hardware entirely