Paper reproduction bundle for 'Prospective difficulty targeting in LLM-generated medical MCQs: a field test of feature-based specification'

Reproduction bundle for the manuscript "Prospective difficulty targeting in LLM-generated medical MCQs: a field test of feature-based specification". Contents. Pipeline source code (code/), the 18 stored item artefacts (cognitive models, generated MCQs, validation reports, metadata), the item register recording source, condition and question type for all 45 examination positions, with intended difficulty for the 18 pipeline-generated items (data/item_register.csv), the aggregated official item statistics from the administered examination, the full generation and validation prompts, rendered figures, and analysis scripts that reproduce Figures 1 and 2, Supplementary Table S5, and every numerical result reported in the paper's Results section. Licensing. The aggregate deposit is released under CC-BY-4.0. The code/ subtree is separately released under the MIT License; the data/ subtree is CC-BY-4.0. See LICENSE_CODE and LICENSE_DATA inside the bundle for full texts. Not included. The administered examination paper and the source USMLE-style respiratory item pool from which the 27 human-authored comparator items were drawn are institutional property and not distributed here; both are available from the authors on reasonable request under a data-sharing agreement. How to cite. Please cite the manuscript as the primary reference and this deposit via its DOI.