We apply Geometric Coupling Topology (GCT) to 1, 867 consecutive letter pairs from Darwin's scientific correspondence (1843–1882) and 497 pairs from two control dyads (publisher, cousin). Using the refraction vector B = Eᵣesponse - Equestion computed on 4096-dimensional text embeddings, we initially report seven findings — and then systematically dismantle all of them through four rounds of internal review, control dyads, and cross-model validation. What does survive is a stylometric gradient: Darwin's directional variability scales monotonically with the intellectual demand of the relationship (Fox g = 0. 69, Murray g = 0. 47, Huxley g = 0. 37, Hooker g = 0. 16, Gray g = 0. 10). The paper contributes a diagnostic checklist of seven confounds and a design principle: GCT measures the dominant variance source between question and answer, and that source is determined by what the experimental design leaves free to vary.
Hanno Koerber (Sat,) studied this question.