Torsion-bar springs remain the dominant primary suspension element in tracked armored platforms of the 30–50 t class — CV90, Puma, K21, Namer, K9 — because they deliver the highest strain-energy density per unit mass of any single-component spring within the geometric constraints of an armored hull (Gomes et al., 2024a). Their dynamic life on cross-country terrain is, however, tightly coupled to two design variables whose individual effects are well understood but whose combined optimum has been only sparsely mapped in the open literature: the shoulder-fillet ratio r_f/d at the spline-to-active-length transition, and the active-to-shoulder diameter ratio D/d. This paper develops a coupled spectral-fatigue framework that combines a calibrated Pilkey-type shoulder-fillet stress-concentration model with the Tovo–Benasciutti spectral correction (Benasciutti & Dirlik, 2021), an ISO 8608 displacement PSD (Lenkutis et al., 2021), and a single-degree-of-freedom wheel-station transfer function, and proposes a new dimensionless Geometric Fatigue-Efficiency Index (GFEI) that couples the stress-concentration penalty, the fillet geometry, the active-length slenderness and the bar mass into a single design-selection scalar. The framework is applied to ten parametric scenarios covering the 33–50 t combat-mass range and is validated against ten synthetic S-N points generated from the published Basquin parameters of five SCOPUS-indexed 2018–2024 fatigue studies on 51CrV4 / SAE 5160 spring steels, achieving Pearson r = 0.580, RMSE = 0.73 decades in log N, and a regression slope of 1.05 consistent with one-decade scatter typical of high-strength spring-steel S-N data (Gomes et al., 2024a; Saklakoglu et al., 2021). Results show that increasing r_f/d from the common-practice value of 0.05 to an optimum of 0.12–0.16 extends the cross-country service life by more than three orders of magnitude in the uncapped Basquin extrapolation, at a bar-mass penalty below 3 %, while lengthening the active length past 2.1 m yields diminishing GFEI gains below 5 % beyond the baseline. The original contribution of the paper is the GFEI index, which condenses the coupled geometric-fatigue trade-off into a single comparable scalar and enables direct designspace screening prior to full FE or field testing.