# SYN-03 — Synthetic Study for Stage 2 Practice

> TRAINING DOCUMENT — This is a fictional study created for Notebook analysis practice. It is not a real publication and must not be cited as research evidence.

## Record information
Study ID: SYN-03  
Year: 2025 (fictional)  
Title: Student experiences with AI feedback in introductory programming labs  

## Abstract
This fictional mixed-method study explored student experiences with AI-generated feedback in programming labs at two universities. Twenty-four students used a feedback assistant for four lab sessions and then completed interviews. Students reported that explanations helped them understand unfamiliar error messages, especially when the assistant referred to their submitted code. They also described checking important suggestions with a lecturer or peer. The study did not include a comparison group or measure programming performance. The authors note that a short trial cannot establish learning effectiveness.

## 1. Introduction
AI feedback may change how novice programmers seek help. This study focused on what students perceived as useful or difficult when receiving feedback during lab work.

## 2. Method
Twenty-four students from two universities participated in four programming lab sessions. The assistant generated an explanation and one suggested next step after students submitted a question and a code excerpt. Researchers conducted semi-structured interviews after the final session and analysed responses using a thematic approach. The report does not provide a detailed coding framework or inter-coder agreement statistic. No control group or pre-test/post-test was used.

## 3. Findings
Three themes were reported. First, students valued explanations that referred to the code they had submitted. Second, students said that generic responses were less useful when a program had several interacting errors. Third, many students described checking high-impact suggestions with a lecturer, teaching assistant or peer before changing their code. The report gives illustrative quotations but does not state how many participants contributed to each theme. Programming performance and course grades were not measured.

## 4. Limitations
The study had 24 participants and covered four lab sessions. It relied on interviews and did not measure learning outcomes. The coding process is only briefly described. Participants came from two universities, but the report does not compare the sites separately.

## 5. Conclusion
The fictional authors report that students valued contextual explanations and human confirmation. They state that the findings describe perceptions and should not be interpreted as evidence that AI feedback improves programming performance.