Structure Follows Causality: Causally-Aligned Representation Separation Breaks Orthogonal Bullying in Multi-Task Learning

When tasks form a causal cascade in multi-task learning—the upstream output conditions the downstream task—one shared representation must encode both the causal parent mechanism and the causal child mechanism, which in the regime we characterize (SGD-class optimizers, sparse high-dimensional downstream, incoherent coupling matrices) occupy incompatible subspaces: task gradients are near-orthogonal ($\cos {\theta }_g\approx 0$, ${\theta }_g:=\angle ({\mathrm{tg}}_1,{\mathrm{tg}}_2)$) and wildly imbalanced ($\Vert {\mathrm{tg}}_1\Vert /\Vert {\mathrm{tg}}_2\Vert \gg 1$), so the dominant task's updates repeatedly erase the weaker task's learning as the latter approaches its optimum—whose pure-curvature form ($E>1$) we verify constructively as a transient rather than asymptotic window. We call this orthogonal bullying and prove it is structural, not algorithmic: the flat joint objective has no stationary point in the persistent excitation region, bullying is unavoidable under testable conditions on dominance, curvature coupling and the crossover threshold, and post-hoc mediators face a detection blind spot or a mediation trilemma (Thms. 1, 2, 3a/3b). We advance Structure Follows Causality—the computational graph must mirror the task causal graph—realized by the Causal Graph Architecture (CGA): each DAG node lives in its own module and conflicting gradient paths are cut by stop-gradient (Props. 4/4${}^{\prime }$, edgewise on any DAG; one line of .detach()), compiled deterministically from any causal DAG (Alg. 1). Under a symmetric comparison protocol—the flat baseline is jointly trained and must satisfy a protection constraint that ${\text{cT}}_2$'s loss never increases at any step, without which flat ${\text{cT}}_2$ fails outright and the rate comparison is vacuous—CGA reduces the iteration complexity of reaching upstream accuracy $\epsilon$ from $\mathcal{O}({\kappa }_g^2\log (1/\epsilon ))$ to $\mathcal{O}(\log (1/\epsilon ))$ (Thm. 5); a cascade-gap decomposition (Thm. 6) shows CGA shares the flat model's information-theoretic floor—its value is representation separation of causal mechanisms, not a spurious statistical gain. On controlled synthetic SCMs CGA holds the single-task floor with near-zero variance (T2 0.10–0.12, T1 0.13; std$\le$0.003; 12 configs, 3 seeds), while Flat fails in mechanism-specific segments; on the v2 legacy $\kappa$-grid CAGrad's upstream error crashes to 71.9–87.0% and GradNorm's to 53.3% (Flat and CGA stay $\le$1.1%). Under strong coupling the system enters a first-order coupling protection regime where bullying does not activate, bounding the assumptions' scope; Flat still pays a 1.7–2.1$\times$ tax and CGA matches the floor. On a production 270-station metro system (Adam + global clipping), joint-training logs record the signature—gradient cosine pinned at 0.00, ${\kappa }_g=200$–$3000$, failed mediation, WAPE 36.6%$\to$37.7%—registered as motivational log only; not a mechanism check, since the production Adam+clipping setting is not reproduced by either controlled arm; the separated flow predictor reaches 10.34% WAPE (Stage-1 trained alone). A pre-registered three-arm exploratory comparison on real large-scale OD data under an identical anti-collapse recipe (single seed; multi-seed confirmation pending) gives flat joint 62.1%, single-task OD without flow 61.2%, CGA 42.7%—a 19.4pp gap indicating joint training cancels the parent's information contribution, preliminary evidence that only causal structure converts it into realizable gain. Further controlled experiments close three gaps: pure-curvature erasure is realized constructively (81/81 single-step, 27/81 sustained) but is transient and unreachable in the MLP family; eight alternative flat protocols all fail to reproduce CGA's floor; and structured sharing (MMoE/PLE/AdaShare) relieves the downstream only by transferring cost upstream (T1 0.300 vs CGA 0.13), leaving CGA the only method on the floor for both tasks. We also give a three-part decision procedure (directionality $\wedge$ $\hat{\Gamma }$ $\wedge$ bullying indicators). Boundaries and negative results are reported honestly: the controlled ${\kappa }_g$ stays below production scale, and the erasure window is transient.