Runbooks work better as decision trees
Linear checklists stall under pressure. Branching decisions match how incidents actually unfold.
A runbook that reads like a novel fails the first time an engineer is tired and the primary dashboard is wrong. Decision trees force clarity: what signal confirms the hypothesis, what to try next, and when to escalate.
Each branch should name a concrete check: a query, a console path, or a known feature flag. Vague steps such as “verify the database” slow response and invite improvisation that is hard to review later.
Keep trees short. If a branch needs more than four steps, split it into a linked procedure. On-call attention is finite; long documents become unread documents.
After every major incident, update the tree that was used. If the team improvised a successful path, capture it. If a branch led nowhere, prune it. Living runbooks beat polished archives.