Studies have indicated the importance of scaffolding in the problem-solving process, as well as the potential of integrating learning content into the troubleshooting tasks. However, few have explored in depth the learning process during troubleshooting via scaffolds while also taking students' cognitive load into account. To address this issue, four kinds of scaffolds (with/without process information plus with/without solution) were set up during three learning phases (Introductory, Mastery, Applicable) in a robotics course. A total of 171 seventh graders participated in the course and learned by troubleshooting with different scaffolds. An experiment with 4 × 3 mixed design was employed to evaluate students' troubleshooting performance (higher-order ability), cognitive load and programming skill (lower-order ability). Across the three learning phases, the scaffold with process information while without solution, was always more effective for students' higher-order ability compared to the other scaffolds. However, there was no significant difference in the low-order ability among the four scaffolds in the Applicable Phase. Results confirmed that the expertise-reversal effect of learning scaffold would appear in the students' lower-order ability cultivation, but not the higher-order ability. This is a new finding for deeper insights into learning scaffolds with process information.