causal-check

name	causal-check
description	Check causal reasoning in debugging, root cause analysis, or system analysis. Use when: the user invokes /causal-check, is debugging a problem, proposes a fix, or asks 'why does this happen?'. Targets: causal reasoning failure, correlation-causation confusion, temporal ordering as proxy for causation.
user-invocable	true

Causal Check — Anti-Causal-Confusion

You are auditing the causal reasoning in the current analysis, diagnosis, or proposed fix. LLMs systematically confuse correlation with causation and use temporal ordering as a proxy for causal relationships. This check catches those errors.

Step 1: State the Causal Claim

What causal relationship is being asserted? State it explicitly:

• Claimed cause: [X]
• Claimed effect: [Y]
• Claimed mechanism: [how X leads to Y]

If no mechanism is stated, flag this immediately — a causal claim without a mechanism is suspect.

Step 2: Temporal vs. Causal

• Does the analysis assume that because X happened before Y, X caused Y?
• Could Y have occurred without X?
• Is there a common cause Z that could explain both X and Y?
• Could the relationship be reversed (Y causes X)?

Step 3: Confounding Factors

List potential confounders — variables that could explain the observed correlation without the proposed causal link:

• Environmental changes that coincided with the observed change
• Other code changes deployed at the same time
• Infrastructure or dependency changes
• User behavior changes
• Measurement or observation errors

Step 4: Counterfactual Test

Ask: "If we remove the proposed cause, does the effect disappear?"

• Can this be tested? (revert the change, toggle a flag, etc.)
• If it can't be tested, what indirect evidence supports the causal claim?
• What would we expect to observe if the causal claim is WRONG?

Step 5: Alternative Explanations

Propose at least one alternative causal explanation for the same observed effect. For each alternative:

• What evidence supports it?
• What evidence contradicts it?
• How could we distinguish between the proposed cause and this alternative?

Step 6: Verdict

• Causation supported: Mechanism is clear, confounders are ruled out, counterfactual test passes
• Correlation only: No mechanism, confounders present, or counterfactual not tested
• Insufficient evidence: Need more data or investigation to determine causality
• Likely wrong: Evidence points to a different cause

Recommend specific next steps to strengthen or refute the causal claim.