<p>The performance evaluation of computer numerical control (CNC) machine tools is constrained by high costs, experimental difficulties, and limited data availability, where small sample sizes further exacerbate the uncertainty in modeling and prediction. An executable high-fidelity digital twin enables effective performance evaluation by bridging physical and virtual environments. Current research rarely focuses on the construction of high-fidelity digital twins via simplified modeling methods or on the development of quantitative metrics for fidelity verification. Therefore, this paper proposes a novel methodology for digital twin construction and introduces fidelity evaluation metrics based on a three-layer meta-action theory. First, the modeling process is simplified by dividing CNC machine tools into system level, chain level, and unit level based on meta-action theory, which improves structural clarity. Then, a digital twin architecture is developed comprising the physical entity, information assurance layer, virtual entity, and functional application layer. A multi-level fidelity assessment framework is proposed with general, system-level, and unit-level indicators. By integrating AHP and the entropy weighting method, the approach enables multi-criteria evaluation and improves the fidelity and reliability of the digital twins. Finally, a case study on axial stiffness performance of a CNC rotary table validates the proposed method, demonstrating its effectiveness in simplifying twin construction and enhancing fidelity evaluation.</p>