Construct validity of dietary mouse models of lean NAFLD

This project documents the data, coding manual, instruments and analysis code for a meta-research study of construct validity in preclinical models of lean NAFLD. We asked whether the dietary mouse models used in this literature (MCD, HFD, CDAA and their combinations) actually reproduce the phenotype on which the published conclusions rest, and whether studies validate the model they use. 300 full-text reports were coded with an LLM-assisted first pass and 100% human verification; a 60-report stratified sample was double-coded blind to estimate reliability. The deposit contains the coded dataset, the final coding manual (v2), the blind-coding instrument, the adjudication workbook, all analysis code, and Figures 1–4 in English and Chinese. This is not a preregistration. The corpus was assembled during a systematic review registered on PROSPERO (CRD420261539779), which was discontinued at title–abstract screening with no outcome data extracted and no synthesis performed. Coding for the present study was completed before this project was made public.