# SYN-02 — 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-02  
Year: 2023 (fictional)  
Title: Automated conversational feedback in a first programming course  

## Abstract
This fictional study examined a conversational AI assistant used by 62 undergraduates in a first programming course over eight weeks. Students could ask the assistant to explain compiler messages and discuss code. The report found no statistically significant difference in final assignment scores between students who used the assistant frequently and those who used it less often. Interviews with 12 volunteers found that students valued quick explanations but sometimes received suggestions that were irrelevant to their code. The authors state that usage was self-selected, so the groups may have differed before the course.

## 1. Introduction
Conversational assistants may provide immediate explanations when teaching staff are unavailable. This report investigated how students used one assistant and whether usage was associated with course outcomes.

## 2. Method
The study involved 62 undergraduates enrolled in one first programming course. Students had optional access to a conversational AI assistant for eight weeks. Researchers grouped students by usage frequency after the course: frequent use (n=28) and occasional use (n=34). Because use was optional, assignment to groups was not random. Final assignment marks and course completion were compared. Twelve volunteers joined semi-structured interviews. The report does not provide the interview guide.

## 3. Findings
The mean final assignment mark was 74/100 for frequent users and 72/100 for occasional users. The difference was not statistically significant. Course completion was 25/28 in the frequent-use group and 31/34 in the occasional-use group. Interview participants described quick explanations as useful when interpreting compiler messages. Five of the 12 interviewees recalled at least one response that did not fit their code. The report does not quantify how often irrelevant suggestions occurred across all users.

## 4. Limitations
Usage groups were self-selected and may have differed in prior experience, motivation or confidence. The study took place in one course. Interview participants were volunteers. The report does not evaluate the correctness of every AI response or measure changes in programming skill over time.

## 5. Conclusion
The fictional authors conclude that students valued immediate explanations, but frequent use was not associated with a statistically significant improvement in final assignment marks. They recommend evaluating response correctness and learner characteristics in future studies.