TeXFix-Bench: Benchmark, Fault Taxonomy, and Campaign Data

Research data package for TeXFix-Bench, an empirically grounded multi-format benchmark for LLM-based document source repair: 10,437 mutation instances from 743 openly licensed seeds; 48,651 hosted repair attempts with engine-gate results; restoration scores; 16k truncation ablation; the mined 18-category LaTeX fault taxonomy with annotation data; and the three-model cross-benchmark. Harness scripts included. API keys and authorization headers excluded.