Operators in intelligent supervisory control tasks need to collaborate and interact effectively in multiple cognitive stages of information acquisition, information analysis, decision making and implementation. However, there is currently a lack of Performance Shaping Factor (PSF) evaluations focused on the cognitive stages, which prevents the analysis of the specific impact of PSFs on human-computer collaboration tasks at a more granular task level. The study invited 19 senior engineers working in the nuclear power control industry to participate in interviews. The interviews focused on defining intelligent supervisory control tasks and exploring the potential impact of 14 PSFs related to human-computer collaboration across different cognitive stages. Based on question coding, we captured a set of task failure probabilities influenced by each PSF. This included the impact of the PSF on individual tasks as well as its specific effects across the four cognitive stages. Finally, the study collected 5320 data points to evaluate the PSFs and utilized Bayesian networks to predict their impact on the failure probability of intelligent supervisory control tasks.