Test predictive validity for generative-AI-assisted coding

Determine whether keystroke-derived editing features retain predictive value for identifying Breakthrough and Fully Stuck students in cohorts that use generative AI coding assistants.

Background

The analyzed CodeBench data were collected before widespread use of generative AI coding assistants. Such tools may substantially alter editing behavior, for example by causing students to insert generated code and perform relatively few subsequent edits. Consequently, the predictive validity of the keystroke-level features identified in the study has not been established for contemporary cohorts using these tools.

References

The predictive value of keystroke-derived features in recent cohorts has not been tested.

— Predicting Struggling Students in CS1 Programming Using Keystroke-Level Editing Features  (2608.25769 - Kofune et al., 26 Aug 2026) in Section 5, Discussion, subsection “Limitations,” paragraph “Generative AI-assisted coding”