A human T cell keeps the shape of its control architecture while swapping out who is in control.

A tiny minority broadcasts nearly all the causal signal (out-degree Gini 0.92; top 5% → ~78% of edges), while a typical gene receives regulation from a moderate number of inputs (in-degree Gini 0.35). Control is broadcast-concentrated but reception-distributed — the first causal confirmation of the Barton/Pritchard 2026 topology prediction, which was made from heritability alone and never tested on perturbation data.

Total edges swing ±32% and up to 59% of the top-100 hubs are replaced across Rest→8h→48h, yet hub-dominance stays pinned at Gini ≈ 0.92. The TCR signalosome (CD3, ZAP70, LAT…) switches on as the activation-state broadcaster — measured only among knockdowns validated in both states, so it is genuine rewiring, not a detectability artifact.

To test whether this is T-cell-specific, we ran the same analysis on a second, fully independent atlas — Replogle 2022 Perturb-seq. The hub-dominated, broadcast-concentrated architecture appears again in K562 (a different cell type, lab and platform), and across K562↔RPE1 the shape holds while hub identity turns over ~80% — so the architecture is a general property of causal gene networks, not a T-cell quirk. Its dynamic shape-invariance now stands as a sharp, falsifiable prediction for the next activation-timecourse atlas — the opening experiment of a full program.

Our turnover axis lets us ask what the atlas paper did not: do the hubs that switch on specifically with activation carry more disease risk than the stable ones? Holding hubness constant, activation-gained hubs are ~2× enriched for monogenic-disease (ClinVar) genes vs stable hubs (OR 2.3, p = 0.01); common-variant autoimmune-GWAS is not enriched, so the signal is specific to monogenic immune genes. The druggable, disease-linked gained hubs — ZAP70, ITK, LCK, PTPRC, IL12RB2 — are candidate state-specific control points: a testable disease hypothesis, not a description.