High-Homogeneity Corpus Hallucination Battery — Synthetic Corpus and Prompt Protocol

Synthetic corpus and experimental protocol for a pre-registered study investigating format-class effects on LLM retrieval failure modes in high-homogeneity structured data. Contains six format class renderings of a 5000-record synthetic corpus with fixed ground truth of 122 qualifying records, the Python generation script (SEED=42, fully reproducible), and the pre-registered prompt battery. Pre-registration: [https://osf.io/ufphj]. Protocol Amendment 1 filed June 16, 2026: Addition of exploratory Prompt Family 5 (Urgency/Constraint Injection). See Protocol_Amendment_1_PF5.txt for full details.