A runbook is not a document that proves you thought about failure. It is a tool someone uses while a system is misbehaving and their context is expensive.
Start with the decision
The first line should tell the reader what they are deciding: “Is this a customer-facing incident?”, “Can we fail over safely?”, “Should we roll back now?” Procedures come after the decision, not before it.
Write for the worst reader
Assume the reader is tired, interrupted, and has never seen this system. Put the exact command, the expected output, and the stop condition next to each other. Avoid “check the logs” unless you name the log, the query, and the time window.
Test the rollback
A runbook that has never been exercised is a hypothesis. Once a quarter, run the rollback path in a safe environment and time it. If it takes longer than the incident it is meant to resolve, the runbook is not finished.