acaus-assume-causation

name	acaus - Assume Causation
description	Force the assumption that X actually causes Y (not just correlates). Trace the mechanism, test the claim, and identify what must be true for causation to hold.
output	{"format":"prose"}

Assume Causation

Input: $ARGUMENTS

---

Core Move

Take a suspected causal relationship and assume it's real causation, not correlation, coincidence, or confounding. Then trace what that requires and what it predicts.

The most common reasoning error is mistaking correlation for causation. This skill forces you to take causation seriously and see what it demands.

---

Procedure

Step 1: State the Causal Claim

CAUSE (X): [what supposedly causes]
EFFECT (Y): [what supposedly results]
CLAIM: X causes Y

Step 2: Force the Assumption

"X genuinely causes Y. The causal arrow runs from X to Y."

Step 3: Trace Causal Requirements

If X causes Y, then ALL of these must hold (Bradford Hill criteria, adapted):

Criterion	Assessment
Mechanism	How does X produce Y? What's the pathway?
Temporal order	Does X precede Y? Always?
Dose-response	More X → more Y?
Consistency	Does X→Y hold across different contexts?
Specificity	Does X cause Y specifically, or everything?
Strength	How strong is the association?
Coherence	Does X→Y fit with what else we know?
Experiment	Manipulating X changes Y?
Reversibility	Removing X removes Y?

Step 4: Test Alternative Explanations

If NOT causation, what else?

1. Reverse causation: Y causes X
2. Common cause: Z causes both X and Y
3. Coincidence: No real connection
4. Mediation: X → Z → Y (X doesn't directly cause Y)
5. Selection bias: We only see X and Y together due to how we're looking

Step 5: Synthesize

CAUSAL CLAIM: X causes Y
ASSUMING CAUSATION:
  Mechanism: [pathway]
  Strongest evidence: [which criteria met]
  Weakest point: [which criteria fail]
  Best alternative explanation: [most plausible non-causal account]
CAUSATION CONFIDENCE: [high/medium/low]
DECISIVE TEST: [what would settle it]

---

When to Use

• Testing whether a correlation is actually causal
• Designing interventions (need real causation, not correlation)
• Evaluating claims of the form "X causes Y"

Integration

• Pair with /arel for general relationship analysis
• Follow with /ht to design a causal test
• Use /aop to assume the opposite (X does NOT cause Y)