Organizational lab simulation with LLM

This dataset contains anonymized written reports produced by undergraduate students enrolled in a Work Organization course in Industrial Engineering. The data were collected across three instructional iterations of a simulation-based learning intervention in which students interacted with a Large Language Model (LLM) agent configured on the ChatGPT platform (model GPT-4) to simulate an organizational context in the sugarcane-energy sector.
The dataset supports the findings reported in an article (being reviewed) which investigates how the pedagogical integration of LLMs, when supported by deliberate instructional design, contributes to learning effectiveness through the lens of Cognitive Load Theory (CLT).
Dataset structure:
The dataset is organized in a single plain-text file (.txt) consolidating all three activity cycles, with each participant's submission clearly delimited by activity label, participant code, and file format (PDF or DOCX). All personally identifiable information — including student names and institutional ID numbers — has been replaced by anonymized codes ([anonymized] and [id_removed]) in compliance with ethical research requirements.
Activity 1 – Group Simulation (in-class, 50 min): 23 digital submissions. Students worked in groups with low scaffolding, producing an interview guide and an initial organizational diagnosis of the simulated company. One additional submission was delivered in physical (handwritten) format and is not included in this repository. Total submissions reported in the associated article: 24.
Activity 2 – Individual Consultancy Simulation (in-class, 80 min): 68 submissions. Students worked individually as consultants, producing structured field-visit reports grounded in Lean Production concepts, supported by high scaffolding and instructor mediation. The associated article analyzed 67 submissions; the additional file corresponds to a late submission not included in the study's performance analysis.
Activity 3 – Individual Extended Diagnostic Analysis (out-of-class, 1 week): 64 submissions. Students produced autonomous diagnostic reports integrating multiple production models (Taylorism, Fordism, Lean, Sociotechnical), with medium scaffolding and no in-class mediation. The associated article analyzed 63 submissions; the additional file corresponds to a late submission not included in the study's performance analysis.
Total submissions: 155 digital anonymized written reports across three activities.