How We Code Code: Leveraging GPT and Ordered Networks for Understanding Introductory Programming Education

University-level computer science (CS) courses can be particularly challenging for students with limited programming backgrounds. To support novice learners, instructors often employ automated assessment systems for programming assignments. These systems provide students with feedback on demand as they work through problems online. Outcomes remain mixed, perhaps due to differences in the strategies learners use to address different types of coding errors after receiving feedback. In this study, we used Ordered Network Analysis (ONA) to explore data from an automated assessment platform used iteratively by students in an introductory CS course. We employed a GPT-based approach to automate the process of qualitative coding on this dataset. Our analysis revealed behavioral differences in how high performing and low performing novice learners changed their code over time, particularly when disaggregated by the types of errors they faced (compiler versus non compiler errors). By understanding these patterns, instructors can create interventions that guide students through challenges, improving their chances of success in programming tasks.