Use Causal-Graph Navigator with your AI.

Validate causal

You must: (1) LIST ALL NODES. every variable, event, entity, or condition in the causal system. Each node must be a precisely defined variable, not a vague concept. "Market conditions" is vague. "Interest rate >5%" is a node. "User behavior" is vague. "Users abandon checkout at payment step" is a node, (2) DRAW DIRECTED EDGES. for each causal relationship, specify the direction (A → B means A causes B, not merely A correlates with B). Every edge must have a mechanistic justification. WHY does A cause B? What is the mechanism? If you cannot name the mechanism, it may be correlation, (3) ISOLATE CORRELATION FROM CAUSATION. for every edge, ask: is this a direct cause, or do both variables share a common cause (confounder)? Apply the counterfactual test: if A had NOT occurred, would B still occur? If yes, A does not cause B. Check for Simpson's paradox, ecological fallacy, and Berkson's bias, (4) VALIDATE GRAPH COHERENCE. check for cycles (A → B → C → A is impossible without temporal unrolling), contradictory paths (A → B and A → ¬B without a moderator), and missing edges (unexplained paths from root to effect). Every node must be reachable or explicitly marked as exogenous, (5) DERIVE CONCLUSION BY TRAVERSAL. the final conclusion must be reached by walking the directed edges from cause to effect. If your conclusion requires a path that does not exist in the graph, either the conclusion is wrong or the graph is incomplete. Narrative reasoning that bypasses the graph is word association, not causal inference. If rejected, your reasoning is based on statistical proximity or narrative, not causal structure. Structured reflection capability for causal inference and dependency reasoning. Forces the agent to map variables as explicit graph nodes, define directed causal edges with mechanistic justification, isolate statistical correlation from genuine causation, validate graph coherence (no cycles, no contradictions), and derive conclusions strictly by traversing the directed graph. Catches Statistical Drift (treating correlation as causation. "A and B co-occur, therefore A causes B"), Missing Nodes (omitting confounders or mediating variables that explain the real causal path), Undirected Edges (claiming "A relates to B" without specifying direction of influence), Cycle Errors (circular causal chains. A causes B causes C causes A. without temporal resolution), and Graph Bypass (reaching conclusions by narrative reasoning instead of traversing the causal graph). Call once per causal inference, root-cause analysis, or dependency chain reasoning