Physical exercise may be associated with less problematic digital use, but the evidence is heterogeneous, and the interpersonal and self-regulatory correlates of this association remain uncertain. We examined the observed-score relationships among exercise, loneliness, self-control, and mobile social media dependence in 205 non-physical-education undergraduates in Jiangxi, China. Item-level questionnaire data were rescored using published keys. Exercise was associated with lower dependence (r = −0.823), lower loneliness (r = −0.799), and higher self-control (r = 0.723; all p < 0.001). However, loneliness and dependence remained extremely highly correlated, and a targeted diagnostic CFA estimated a boundary-level latent correlation of approximately 0.999, indicating inadequate discriminant validity. An exploratory PROCESS-equivalent Model 15 showed significant exercise × self-control (B = 0.0133, p < 0.001) and loneliness × self-control (B = 1.0042, p < 0.001) interactions; the conditional product-of-coefficients index was −0.0177, 95% bootstrap CI [−0.0244, −0.0113]. HC3 and influential-case sensitivity analyses preserved the interaction pattern. The data therefore support a robust observed-score association between greater exercise and lower dependence, but they do not establish a distinct loneliness-mediated mechanism or causal ordering. These findings highlight the importance of integrating behavioral, interpersonal, and self-regulatory perspectives while prioritizing construct differentiation and measurement validity in research on university students’ digital well-being.