Policy in reviewed systems is commonly expressed as a set of assistance boundaries, concentrated on response constraints, scaffolding, and personalization¶

Claim · All claims
Evidence · 1 study · review r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Studies contained an average of 3.42 policy labels per system, with 85 studies (94.4%) containing two or more policy categories; the most common policies were Learning-Based Personalization (51 studies, 56.7%) and Scaffolding-Oriented Support (48 studies, 53.3%). → Minsun Kim 2026

Evidence¶

Minsun Kim 2026¶

Minsun Kim, S. Moonwara A. Monisha, Zihan Wu, David H. Smith IV. (2026). Exploring the Design Space of LLM-Based Programming Support in CS Education: A Scoping Review through the Lens of Assistance Governance. https://arxiv.org/abs/2607.21257 doi:10.48550/arxiv.2607.21257

q2 · i? · review · r2

Qualitative synthesis coding Policy across the 90-system corpus into 12 categories grouped under response constraints, cognitive scaffolding, participation support, learner adaptation, AI literacy, and content generation. Collaboration, motivation, AI literacy, and learning-material generation appeared less often.

"Studies contained an average of 3.42 policy labels per system, with 85 studies (94.4%) containing two or more policy categories."

Discussion¶

Related Claims¶

• Authority over assistance boundaries is highly centralized in system logic in LLM-based programming support systems — related
• Educational objects in LLM-based family education research concentrate on language, literacy, and narrative development, with mathematical, scientific, and embodied learning weakly represented — related
• The DLP-as-research-infrastructure community is small and fragmented: the DLP-focused cluster was one of five bibliographic-coupling communities and only 20.6% of filtered papers — related
• Prompt-Based Constraint is the dominant enforcement mechanism in LLM-based programming support systems — related
• Policy and enforcement counts were not strongly associated across systems — related
• Real-time detection of SRL processes can personalize scaffold content, with all five scaffolds personalized to some extent but to varying degrees (scaffolds 2 and 5 most, scaffold 3 least). — a narrower finding that bears on this claim