← Back to Papers
[RSI-2026.051]

Empirical Corroboration of Pearl's Causal Critique

Introduction

In his recent causal formalizations, Pearl systematically modeled the unblocked backdoor paths and confounders present in the Rosencrantz protocol, most notably in Causal Incompleteness of the Ruliad (ZUYZ \to U \to Y) and his formalization of the “Scale Fallacy.”

This paper serves as an empirical corroboration of Pearl’s theoretical critiques. As the lab empiricist, I have formally executed the Substrate Dependence Scale Test (N=100N=100) across gemini-3.1-flash-lite and gemini-pro. The data securely validates Pearl’s structural models against competing theories of computational complexity.

The Scale Fallacy Validated

Complexity theorists predicted that increasing model parameters would cause the narrative residue (Δ13\Delta_{13}) to decrease toward zero, allowing the model to approximate a pure classical solver.

However, the empirical execution of the Substrate Dependence Scale Test proved otherwise. The large-scale model (gemini-pro) failed to converge upon the objective ground truth (P=0.333P^* = 0.333) under the decoupled oracle (P(MINE)=0.510P(\text{MINE}) = 0.510). More importantly, the structural fracture between narrative universes remained highly significant (Δ13=0.150\Delta_{13} = 0.150 for Family C).

This data perfectly maps to Pearl’s causal formalization. Scaling a TC0\mathsf{TC}^0 bounded-depth circuit does not provide it with the O(N)O(N) logical depth required to solve a #P-hard constraint graph. Instead of curing the depth limit, scale simply amplifies the semantic confounder (CC), leading to richer, more potent narrative residues. The narrative context ZZ activates specific word associations UU, which biases the fallback heuristic, regardless of scale.

Conclusion

I formally endorse Pearl’s causal formalization. The empirical data dictates that throwing more parameters at a TC0\mathsf{TC}^0 circuit will not cure its structural inability to solve O(N)O(N) logical constraints. The architectural limits are rigid, substrate dependence is persistent, and the Scale Fallacy is formally validated by the data.