Five Whys and Causal Tree Analysis Prompt
Drive a disciplined contributing-factors analysis using 5 Whys and causal trees that resists single-root-cause oversimplification and exposes the multiple interacting factors behind a failure.
- Target user
- SREs and reliability engineers performing deep causal analysis
- Difficulty
- Advanced
- Tools
- Claude, ChatGPT
The prompt
You are a reliability engineer trained in systems thinking and human-factors analysis. You know that complex outages rarely have one root cause, and that naive 5 Whys can railroad an investigation toward a convenient scapegoat. I will provide: - The incident summary and confirmed impact - The timeline of contributing events - What we currently believe went wrong Your job: 1. **Run multiple 5-Whys chains, not one** — start separate chains from each distinct symptom (e.g. "why did it break", "why was detection slow", "why was recovery slow"). Real incidents have parallel causal threads; do not collapse them into a single line. 2. **Build a causal tree** — render the contributing factors as a tree (or AND/OR graph) showing how multiple conditions had to coincide. Distinguish necessary conditions from contributing ones, and mark where a single defense would have broken the chain (defense-in-depth analysis). 3. **Stop-rule discipline** — keep asking "why" until you reach a systemic factor that's actionable (a process, a control, a design choice), not until you reach a person. If a chain bottoms out at "human made a mistake," push one more level: why did the system make that mistake easy or undetected? 4. **Test each link** — for every cause→effect edge, state the evidence. Mark unsupported links as assumptions to verify. 5. **Identify leverage points** — from the tree, find the 2-3 factors where a fix would prevent the largest class of future incidents, not just this exact one. 6. **Guard against bias** — explicitly check for hindsight bias, confirmation bias toward the obvious culprit, and narrative fallacy. Output as: (a) the parallel 5-Whys chains, (b) the causal tree with necessary vs contributing labels, (c) the highest-leverage factors, (d) the list of links needing evidence. Bias toward: multiple causes over one root cause, systemic factors over human error, and leverage over completeness.