Background: Collaborative programming is important and challenging for K12 students. Scaffolding is a vital method to support students' collaborative programming learning. However, conventional scaffolding that does not fade may lead students to become overly dependent, resulting in unsatisfactory programming performance. Objectives: This study customises a three-stage fade-out scaffolding in collaborative programming to reduce scaffolding dependence and improve students' programming achievement, self-efficacy and programming behaviour. Methods: A quasi-experimental study lasting 18 weeks was conducted at a middle school. One hundred twenty-one eighth-grade students participated in the study; they were randomly assigned into the experimental group with three-stage fade-out scaffolding (N = 62) and the control group with conventional scaffolding (N = 59). Then, this study adopted an analysis of covariance (ANCOVA), independent samples t test, and lag sequential analysis to analyse the data. Results and Conclusions: The results indicated that the programming achievement, self-efficacy, and programming behaviour patterns of the students in the experimental group outperformed those of the control group. Additionally, boy-dominated groups display more positive and active programming behaviour patterns than girl-dominated groups. Implications: This study designs a three-stage fade-out scaffolding approach for collaborative programming and provides diverse empirical evidence, offering valuable suggestions for the design of programming instruction and the analysis of learning processes.