National research governance faces a strategic challenge: traditional metrics reward cumulative output over transformative impact, concealing high-risk innovation and reinforcing structural biases. Current data infrastructures offer limited visibility into an increasingly complex, interconnected ecosystem. We propose the Data → Graph → Graph Optimization paradigm as a foundation for a Research Digital Twin - a dynamic, auditable representation of the national research system. This architecture integrates three critical dimensions: social (collaboration networks), semantic (knowledge evolution), economic (resource flows) into a unified temporal knowledge graph. We are building toward this vision through concrete prototypes: semantic impact prediction in biomedical research (moving beyond citation metrics) and talent mobility modeling in research organizations. These demonstrate how LLM-powered systems can transform unstructured data into analyzable graphs, while Retrieval-Augmented Generation ensures AI reasoning remains traceable to verified sources. The full paradigm enables a new class of strategic capabilities: automated hypothesis generation from knowledge gaps, multi-objective optimization across innovation and equity goals, and scenario simulation for policy decisions. Built on open-source infrastructure, this approach ensures methodological transparency and sovereign ownership of research intelligence assets. By translating strategic questions into quantitative, explainable signals, the Research Digital Twin offers decision-makers genuine foresight for steering national research ecosystems toward innovation, fairness, and long-term resilience..