The extent to which working memory (WM) training can produce far transfer to fluid intelligence (Gf) remains a matter of considerable debate. The present paper draws on neuroimaging evidence indicating that individuals with autism spectrum disorder (ASD) show enhanced activation in posterior visual network regions—including the cuneus and lateral occipitotemporal cortex—during performance of Raven's Progressive Matrices (RPM; Soulieres et al., 2009), and proposes a new N-back paradigm, Nexus N-back, designed to preferentially engage these posterior pathways. Whereas conventional Dual N-back relies on audiovisual cross-modal stimulation and primarily recruits frontoparietal networks, the proposed paradigm requires simultaneous tracking of three visual channels—spatial position (dorsal pathway), shape (ventral pathway), and color (area V4)—each selected for its hypothesized overlap with the neural substrates engaged during RPM performance. Three anti-strategy mechanisms are incorporated: N−1 back lure trials, inter-stimulus interval jitter, and accuracy-triggered channel rotation. The present work differs from the exploratory quad N-back approach of Lindeløv et al. (2019) in its neuroscientifically grounded channel selection, deliberate exclusion of auditory stimulation, and structured anti-strategy design. This paper constitutes a theoretical proposal without experimental verification; its aim is to provide a principled design framework for future randomized controlled trials.