Scale dependence is a pervasive feature of socioeconomic networks: even when generated by the same class of economic activities, networks observed at different levels of entity aggregation can exhibit markedly different structural organizations, yet a systematic and testable explanation for such cross-scale divergence remains lacking. This paper presents a U.S. patent dataset together with a unified analytical framework that combines empirical network analysis with mechanism-based modeling to quantify and interpret structural differences across scales. The dataset contains 1,225,373 granted USPTO utility patents filed during 2000—2020 and integrates assignee geography (state/county/city), firm identifiers, CPC classification codes, and patent texts; technologies are defined at the 4-digit CPC level. To measure technology activity when patents involve multiple technologies, we use co-citation information to allocate each patent's technological shares across its associated CPC codes, thereby obtaining technology-share weights beyond naive equal counting. Using these weights, we construct entity—technology bipartite networks at four scales (state, county, city, and firm) and derive two technology space networks, one based on technology co-occurrence across entities and the other based on patent-text similarity. We characterize network structure using bipartite modularity (Q), global nestedness (N), and in-block nestedness (I), and evaluate statistical significance against degree-constrained null models based on the bipartite configuration model (BiCM). Empirically, modularity increases as the entity scale becomes finer; state- and county-level networks are closer to a globally nested organization, whereas city- and firm-level networks exhibit a pronounced shift toward in-block nestedness. Temporal analysis further shows that the formation of new entity—technology links reflects a scale-dependent balance between preference for globally central technologies and reliance on relatedness density to the existing technological portfolio, with smaller-scale entities exhibiting a stronger dependence on relatedness density. Finally, we propose an evolutionary model that incorporates both relatedness-density preference and technology-centrality preference under empirical degree constraints. Simulations demonstrate that tuning these two preferences reproduces the observed transition from global nestedness to in-block nestedness, providing a mechanism-based explanation for scale-dependent structural patterns in technological innovation networks.