"ProgCAS: A Programming Dialogue Dataset for Temporal Cognitive and Affective State Modeling"

"Modeling learners\u2019 evolving cognitive and affective states is essential for learner-centered agentic AI in self-directed programming learning. Yet existing datasets for programming education often emphasize code correctness, submission histories, or static performance outcomes, leaving insufficient resources for studying how learners\u2019 states unfold through human\u2013AI dialogue. We present ProgCAS, a programming dialogue dataset for temporal cognitive and affective state modeling. The dataset consists of multi-turn learner\u2013AI interactions collected during self-directed project-based programming tasks, together with process-level traces related to exploration, help seeking, debugging, revision, and task progression. ProgCAS provides annotations of temporal learner states, including cognitive indicators such as conceptual uncertainty, implementation progress, misconception cues, and reflective understanding, as well as affective indicators such as frustration, confidence, hesitation, and engagement. By aligning dialogue content with learner-state trajectories, ProgCAS enables the study of how AI support can be conditioned on learners\u2019 evolving needs rather than only on immediate task requests. The dataset is intended to support future research on adaptive scaffolding, affect-aware programming support, learner modeling, and educational agentic AI systems that preserve learner agency while supporting project completion."