The HARMOGEN-R dataset of human and AI-generated rubric evaluations for formative programming assessment

HARMOGEN-R Dataset: Large-Scale Human–LLM Evaluation Records for Automated Grading Research The HARMOGEN-R dataset comprises 68,948 structured records of educational assessments comparing human and Large Language Model (LLM) grading. It includes 770 anonymized student responses from a Data Structures and Algorithms course, evaluated by human instructors and several LLMs using five rubric variants: one human-created and four AI-generated. Each response was assessed across multiple criteria, producing 50,050 individual evaluations, 15,400 aggregated scores, and 2,310 evaluation reports. The dataset is organized into 15 relational tables with enforced foreign key constraints and validated JSON fields that document the reasoning process used in automated grading. Each LLM evaluation includes a criteria_evaluations_by_llm field that stores the structured rationale applied by the model to each criterion, including intermediate sub-scores and textual justifications. This structure supports reproducible quantitative and qualitative analyses of model evaluation behaviour. The repository provides the complete dataset in multiple formats, including SQL, per-table CSV exports, and per-table JSON exports, as well as analytical CSV extracts for statistical analysis. It also includes Python scripts for data extraction, normalization, and exploratory analysis. All identifiers are anonymized, and the dataset does not contain personal information. Version 2.0 corrects a data issue in the final_evaluations table by removing 80 records associated with a discarded model configuration. It also adds per-table CSV and JSON exports, a flattened criterion-level CSV file, and an export script to improve accessibility and reproducibility. The HARMOGEN-R dataset supports empirical research on automated assessment, rubric generation, and human–AI grading consistency. It is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.