RQ2: Preciseness of abstract model

We further evaluate the preciseness of the selected abstract model under different scenarios. The results are presented in Table 5, where random, safe, and unsafe refer to analyses performed using randomly sampled, safe, and unsafe traces, respectively. We notice that, across 16 abstract models, the average preciseness over all evaluation traces is 93.44%, with most abstract models achieving above 90% accuracy for both safe and unsafe cases. This high and consistent preciseness across different scenarios indicates that the abstraction retains sufficient expressiveness to capture the robustness semantics of the underlying system behaviors.