Requirement–State Transition Diagram Dataset for LLM Evaluation

This dataset contains 80 software requirement cases paired with UML state transition diagrams represented in PlantUML. It supports research on LLM-based diagram generation, retrieval-augmented generation, automated validation, iterative repair, semantic evaluation, and human and LLM-based quality assessment. Each case includes: an original natural-language requirement; an aligned requirement; a structured requirement specification; a reference PlantUML diagram; and a PNG rendered from the PlantUML source. The cases cover multiple software application domains and varying levels of diagram complexity. The archive also provides a dataset manifest, data dictionary, README, license, and citation metadata. The dataset is intended for software engineering research and education, particularly experiments involving requirement-to-diagram generation, syntactic and structural validation, violation-specific repair, semantic state matching, and LLM-as-a-Judge evaluation.