Research data and reproducibility package for a deterministic-generative framework for auditable AI-assisted water quality monitoring and interpretation
This record contains the research data and reproducibility materials supporting the study “Separating Calculation from Communication in AI-Assisted Water Quality Interpretation”. The archive includes the processed INEMA Monitora dataset used to construct the observational scenarios; experimental prompts and metadata; deterministic structured records; complete outputs from GPT-4.1 mini, Gemini 2.5 Flash, and Qwen2.5-72B Instruct; primary evaluator ratings; the blinded external evaluation; the matched 54-pair analysis matrix; and the Python and R scripts used for statistical analyses and manuscript figures. The materials are provided to support verification, transparency, and reproducibility of the reported analyses. Observational scenarios were derived from publicly available INEMA Monitora data for 2020–2021.