Pedagogical Mediation of Generative AI in Programming Education: Systematic Literature Review Dataset (Anonymized Version)

This repository provides the complete methodological artifacts of a systematic literature review with explanatory synthesis on generative AI (GenAI) in programming education. The review covers 101 peer-reviewed studies published between 2021 and 2025, retrieved from seven digital libraries (ACM Digital Library, DBLP, IEEE Xplore, ScienceDirect, Scopus, the Brazilian Computer Society Digital Library, and SpringerLink) over a search window of 2020 to 2025. The corpus comprises 94 empirical studies and 7 theoretical or conceptual works. The repository includes: Search protocol. The canonical search string, the queries as submitted to each digital library, and the eight anchor studies used to validate the strategy. Screening records. Inclusion and exclusion criteria and the decision recorded for each of the 1,331 retrieved records across three sequential filters (Filter 1, title and abstract; Filter 2, full text; Filter 3, methodological rigor), allowing the PRISMA flow to be reconstructed record by record. Methodological rigor assessment. Assessment records for the 102 studies that reached Filter 3, scored with a ten-item instrument on a 0 to 10 scale, together with publication venue, CORE ranking, SJR band, and year. Data extraction tables. Structured extraction for all 101 included studies, organized by research question and covering study characteristics, GenAI tools, instructional contexts, pedagogical mediation strategies, and reported outcomes, plus a focused extraction for the 13 studies that examine metacognition and self-regulation. Configurational coding. Per-study coding of the predominant outcome direction for all 101 studies, and of the combination of pedagogical mediation, tool design, and metacognitive indicator for the 29 studies with assessable outcomes, each with the supporting textual evidence. The moderators reported in the associated article are candidate moderators, identified through co-occurrence patterns between study characteristics and reported outcomes rather than through statistical testing of moderation. No causal claim is made or supported by these files. These artifacts support full reproducibility of every stage of the review and are made available to enable independent replication and secondary analyses. This is an anonymized version prepared for peer review; the identified version will be released upon publication.