Kim, Monisha, Wu & Smith IV (2026) — Virginia Tech; University of Maine. arXiv preprint (cs.HC). 📄 Full text (arXiv)
This scoping review synthesizes 90 peer-reviewed llm-based programming support systems in cs-education to make explicit how each system bounds, enacts, and controls assistance — decisions the authors argue are usually left implicit. They introduce the PEA framework: Policy (what help is allowed or restricted), Enforcement (how boundaries are operationalized through interaction and system behavior), and Authority (who can configure, adapt, or override them at runtime). The analysis shows systems often share pedagogical goals but diverge sharply in enforcement mechanisms, while authority remains highly centralized in system logic — few tools give learners or instructors runtime control. PEA offers a comparative vocabulary that connects to genai-assessment-governance and genai-policies-higher-ed-computing, and its emphasis on bounded, configurable help echoes structured-llm-feedback-programming and llm-feedback-programming-classroom. The governance lens reframes scaffolding as a design-space question of control rather than a fixed pedagogical feature.
Related Pages
- cs-education — the disciplinary setting for the reviewed systems
- genai-assessment-governance — parallel governance framing for assessment
- genai-policies-higher-ed-computing — policy layer of assistance governance
- structured-llm-feedback-programming — example of bounded programming feedback
- llm-feedback-programming-classroom — classroom deployment of LLM programming support