Model merging in model-based engineering imposes high cognitive demands, as developers must identify, reconcile, and integrate changes across parallel artifacts. To explore the neural dynamics of this process, we conducted a controlled experiment with 35 developers performing merge tasks in two representational formats—sentential (Java code) and diagrammatic (UML class diagrams)—while EEG was continuously recorded. We clustered EEG signals into the four canonical microstate classes (A–D), which are well-established in cognitive neuroscience and represent recurring large-scale brain activity patterns. In alternating cycles of eyes-closed rest and task execution, we analyzed microstate dynamics and found diametrically opposite patterns depending on representation: in the sentential format, microstates A, C, and D increased significantly in duration, occurrence, and coverage, whereas in the diagrammatic format, A and C decreased, B declined moderately, and D showed reductions in occurrence and coverage. These findings reinforce the feasibility of microstate metrics as fine-grained biomarkers of cognitive overload in model-integration operations, paving the way for cognitively aware tools and adaptive environments capable of automatically adjusting abstraction levels or issuing high-load alerts.