Correlation isn't causation. Neither is sequence. "This happened, then that happened" is not the same as "this caused that" — but it's treated as equivalent constantly, and it produces wrong diagnoses, failed fixes, and surprised engineers.
This skill makes causal structure explicit: what actually depends on what, what change produces what effect, and what must be true for a plan to hold.
Four Modes
Use the mode that matches the question.
Framing check: Confirm the specific causal situation before selecting a mode. State what you've identified — the system or situation, the observed effect or proposed change, and the core causal question — in one sentence, then use AskUserQuestion:
- Question: "I'm reading this as: [your one-sentence framing of the specific situation and causal question]. Is that right?"
- Header: "Framing"
- Options:
- Yes — proceed — framing is correct
- Adjust — one element is off; user will correct it before you continue
- Reframe — different situation than read; incorporate the correction before proceeding
Mode 1: Root Cause Tracing
"Why did this happen?"
Work backwards from an observed effect to its cause — and to the cause of that cause.
Process:
- State the observed effect precisely. Not "the system is slow" — "p95 latency increased from 120ms to 840ms after the Tuesday deploy."
- Ask: what are the immediate causes that could produce this effect? List all plausible candidates.
- For each candidate: what evidence would confirm or rule it out?
- Eliminate candidates. For the survivors: what caused them?
- Continue until you reach a cause that has no upstream cause within scope — or a point where further tracing requires different expertise or data.
- Distinguish: root cause (the origin), proximate cause (the immediate trigger), contributing factors (conditions that allowed it).
Mode 2: Impact Mapping
"What breaks if I change X?"
Work forwards from a proposed change through its downstream effects.
Process:
- State the change precisely.
- Identify direct dependents: what immediately relies on the thing being changed?
- For each dependent: what behaviour changes, and what downstream systems rely on that behaviour?
- Continue for two to three levels of dependency.
- Mark irreversible effects — changes that, once made, cannot be cleanly undone.
- Mark cascade risks — places where a small effect triggers a large one.
Mode 3: Dependency Mapping
"What must be true for this to work?"
Identify the full set of conditions a plan depends on.
Process:
- State the plan or goal.
- Ask: what must be true in the environment for this to succeed? List all dependencies.
- For each dependency: is it guaranteed, assumed, or unknown?
- For each assumed dependency: what is the cost if the assumption is wrong?
- Surface single points of failure — dependencies that, if they break, cause total plan failure with no fallback.
Mode 4: Counterfactual Testing
"Would Y still have happened without X?"
Test a causal claim by reasoning about the counterfactual world.
Process:
- State the causal claim: "X caused Y."
- Imagine removing X. Would Y still happen via another path?
- If yes: X is a contributing factor, not the root cause.
- If no: X is necessary for Y — strong causal evidence.
- Consider: was X sufficient alone, or did it require other conditions?
Human Check-in
Before proceeding, use the AskUserQuestion tool. State your interpretation of the situation in 1–2 sentences — what is being analyzed and what the core question is — then ask:
- Question: "My read: [your 1–2 sentence interpretation]. How do you want to proceed?"
- Header: "Scope"
- Options:
- Full analysis — Complete all steps, reasoning shown throughout
- Key findings only — Bottom-line output, skip step-by-step detail
- Direct causes only — Map first-order causal links, skip downstream consequences
- Reframe — The read is off; correct it and the analysis will follow the corrected framing
Proceed based on their selection. If the user reframes, incorporate the correction before running any analysis.
Output Format
Mode: [Root Cause / Impact Mapping / Dependency Mapping / Counterfactual]
Question: [the specific causal question being answered]
[Mode-appropriate structure]
For Root Cause:
- Proximate cause: [immediate trigger]
- Root cause: [origin]
- Contributing factors: [conditions that enabled it]
- Evidence: [what confirms this vs what was ruled out]
For Impact Mapping:
- Direct effects: [level 1 dependencies affected]
- Downstream effects: [level 2-3]
- Irreversible effects: [flagged]
- Cascade risks: [flagged]
For Dependency Mapping:
- Dependencies: [each, with status: guaranteed / assumed / unknown]
- Single points of failure: [flagged]
- Highest-risk assumptions: [ranked]
For Counterfactual:
- Causal claim tested: [X caused Y]
- Counterfactual: [would Y happen without X?]
- Verdict: [necessary / contributing / coincidental]
- Conditions required alongside X: [if X alone wasn't sufficient]
Summary [2-3 sentences on what the causal analysis reveals and what action it implies]
Notes
Causal reasoning is always provisional — it produces the best available model given current evidence, not a proof. State explicitly what evidence would change the analysis. In complex systems, multiple causal chains often contribute to a single effect; resist the urge to stop at the first plausible explanation.
What's Next
After delivering this output, use AskUserQuestion to offer the next move:
- Question: "Causal chain mapped. What's next?"
- Header: "Next"
- Options:
/s4h-systems-feedback-mapping— Turn the causal map into a feedback loop analysis/s4h-historical-precedent-analysis— Check whether this causal chain has played out before/s4h-constraint-hardness-testing— Test which causal link is the weakest and most brittle- Done — Wrap up and synthesise what we have so far