We consider a recently proposed simulation-based decision-making framework, called offline-simulation-online-application (OSOA). In this framework, simulation experiments are not performed after the target problem is set up with all the input parameters; instead, they are performed before that with all the possible parameters that might come up in the target problem. Then, these computational results are directly used in real time for system evaluation or system optimization when the input parameters of the target problem are known. In this paper, we follow this framework and use stochastic kriging (SK) to model the system performance from the covariate space. Two measures, namely IMSE and IPFS, are proposed to evaluate the prediction errors for system evaluation and system optimization respectively. We establish the convergence rates of these two measures. They quantify the magnitude of prediction errors for the online application after a certain period of time is spent for the offline simulation.