In school practice, one obstacle to the spread of computer-based, AI-supported testing is that most tasks still originate with teachers, making task bank maintenance time-consuming. In our previous work, we showed that student-generated questions can be incorporated into the assessment process when teacher experience and student performance data are handled through a fuzzy-based method integrated into the WTCAI (When The Child Asks with AI) school decision support system. This study extends that framework by addressing a specific problem: the difficulty label assigned by the teacher at the time of task creation may become inaccurate as successive cohorts with different profiles solve the same item. We present a fuzzy rule-based dynamic update mechanism that recalibrates task difficulty based on observed class-level performance, without overriding teacher judgment. Drawing on a previously established school dataset and school-based observational data [18], we explore this problem and present a practical, teacher-interpretable algorithmic solution. The paper, therefore, presents a practical and interpretable recalibration framework, with empirical validation currently underway.