Citation-verification audit: reproducibility deposit for "Omit-Then-Verify" (CodeBluePediatrics)

This deposit accompanies the manuscript "Omit-Then-Verify: A Deployed Defense Against Fabricated Citations in LLM-Generated Medical Content" (B. Hsu) and contains only the artifacts needed to independently reproduce the manuscript's central citation-verification audit. In a prospectively specified, fixed-seed random sample of 300 references (298 scorable) drawn from a deployed pediatric clinical-pathway library, 279 of 298 (93.6%; 95% Wilson CI, 90.3% to 95.9%) resolved to a real, topic-matched PubMed record. Files: reference_list.jsonl (the de-identified sampling frame of structured bibliographic entries), results.jsonl (per-reference verification outcomes), verify.py (a standalone script that re-draws the identical seeded sample, queries live PubMed via NCBI E-utilities, and prints the verified rate with a 95% Wilson confidence interval), and README.md (method and caveats). No patient data are present. The CodeBluePediatrics pipeline, prompts, drug database, and clinical-pathway text are proprietary and are not part of this deposit.