Computer simulations play a vital role in cyber-physical systems, facilitating the verification of operational behavior and supporting decision-making under diverse conditions. Over time, various simulation methodologies have been developed across disciplines, spanning the natural sciences, social sciences, and industry. Calibration serves as a crucial preprocessing step to ensure simulations accurately reproduce realistic data. However, repeated simulation runs impose significant computational costs. To mitigate this challenge, the model bridge method has been introduced to establish a relationship between the surrogate model and the simulation, leveraging past calibration results. Experimental validation has demonstrated its efficacy in reducing computational costs. Nevertheless, its accuracy remains uncertain, and quantitative insights into the factors influencing its improvement are lacking. This study conducts an asymptotic analysis of accuracy and elucidates the impact of past datasets and surrogate model properties on the performance of the method.