The global live streaming commerce (LSC) ecosystem is persistently marked by a "conversion paradox," wherein elevated audience engagement fails to materialize into stable transaction volumes. Although the Stimulus-Organism-Response (S-O-R) framework constitutes the dominant theoretical scaffold for impulsive buying research, its empirical performance within the Indonesian market remains volatile, with prior studies yielding contradictory evidence on the efficacy of scarcity messagingĀ a fragmentation attributable, in part, to the systematic neglect of intra-national market heterogeneity. Addressing this gap, the present study introduces geographic regionĀ specifically the Java versus Non-Java dichotomy as an empirically validated structural boundary condition within the S-O-R framework, representing one of the earliest empirical efforts to operationalize intra-national digital market segmentation as a theoretically grounded moderator within the LSC literature. A cross-sectional survey of 370 Shopee Live users (Java: n=247; Non-Java: n=123) was analyzed using PLS-SEM, with measurement equivalence and inter-group structural comparisons rigorously established through MICOM and Permutation-Based MGA. Results reveal substantive regional heterogeneity: in the Java segment, scarcity messages exert both direct and FOMO-mediated effects on impulsive buying, whereas in Non-Java, the influence operates almost exclusively through affective FOMO arousal. MGA confirms statistically significant structural path differences (p < 0.05), demonstrating that regional socio-digital context functions as a substantive moderator of consumer decision heuristics. These findings advance the S-O-R paradigm by repositioning it from a universalist model toward a context-contingent theoretical architecture, with practical mandates for regionally calibrated e-commerce strategies direct scarcity cues in digitally mature markets and narrative-driven FOMO cultivation in emerging segments. The study is bounded by its single-platform scope (Shopee Live), a cross-sectional design that precludes causal inference, and a binary regional classification that, while theoretically grounded, necessarily simplifies Indonesia's broader socio-digital continuum.