Data and code for: Shell records and the limits of metadata: measuring registered versus accessible research data in Australian universities

<p dir="ltr">This deposit contains the data, code and supplementary analysis for the article "Shell records and the limits of metadata: measuring registered versus accessible research data in Australian universities" (submitted to Journal of Data and Information Science, 2026).</p><p dir="ltr">The study is a census of all 17,226 DataCite-registered dataset records across 41 Australian universities, harvested from the DataCite public REST API on 2–3 June 2026.

It examines the scale, disciplinary distribution, licensing and FAIR compliance of these records, with particular attention to records that declare no files (metadata-only or "shell" records), and compares DataCite registrations with the dataset counts repositories advertise.</p><p dir="ltr">Files</p><ul><li>shell_records_data.xlsx - the main workbook. The README tab documents all variables; the other tabs hold the full record-level harvest (all_records), F-UJI FAIR assessment scores for a stratified sample of 705 records (fair_scores), and institution-level summary statistics combined with a manual audit of each university's data repository (institutions).</li><li>institutions_config.csv - the institution-to-DataCite-client configuration file read by harvest.py.</li><li>harvest.py - harvests dataset records from the DataCite public REST API.</li><li>fair.py - runs F-UJI automated FAIR assessments on the sampled records.</li><li>supplementary_analysis.pdf / .docx - supplementary analyses referenced in the article.</li></ul><p dir="ltr">The Python scripts were written with the assistance of Claude Sonnet 4.6 (Anthropic) under the direction of the author(s), who specified, tested and verified them.