<p>This paper develops a Cognitive-Network Framework (CNF) that integrates cognitive theory, observable behavioral manifestations in discourse and interaction, and computational indicators derived from NLP and network analysis. The CNF distinguishes theoretical mechanisms (motivated reasoning, identity protection), their meso-level behavioral signatures (selective sharing, sentiment homogeneity, bridge erosion), and the computational measurements (transformer classification, GNN dynamics, time-series features) that serve as proxy indicators of those signatures. XLM-RoBERTa models are used for multilingual classification and feature extraction, while temporal GNNs and VAR/LSTM architectures are used for dynamic modeling and short-horizon forecasting within validated time-series frameworks. The framework identifies four mechanisms of civic culture degradation—civil society restriction, media capture, opposition harassment, and institutional erosion—and shows how civic-electoral divergence (declining civic culture despite continued electoral competition) emerges from interactions among cognitive, network, and technological processes. The approach combines large-scale behavioral data, survey validation, experimental replication, and qualitative tracing to support inference from computational indicators to broader civic outcomes. Construct validation demonstrates consistent relationships between computational proxies and survey measures (example correlation <i>r</i> = 0.67 for partisan identity strength and within-community sentiment homogeneity), and cross-level interaction models show that cognitive × network effects substantially increase explanatory power for civic culture deterioration.</p>