Month: 2026-03 — Focused feature delivery and code quality improvements in modelscope/data-juicer. Delivered the LaTeX Figure Extraction Mapper Operator, enabling extraction of figures and their citing context from LaTeX sources. The operator outputs one or more rows per figure/subfigure with fields such as images, caption, label, citing_paragraphs, parent_caption, and parent_label, facilitating downstream analytics, indexing, and content reuse. Implemented support for figure environments (figure/figure*), wrapfigure, subfigure/subfloat, and citations via \ref/\cref/\autoref, with a robust recursive nested-brace regex implementation. Delivered comprehensive unit tests (21 cases) and bilingual documentation (EN/CN). Added config entry in config_all.yaml and registered the operator in the mapper __init__.py to enable seamless integration into the data-juicer workflow. This work unlocks end-to-end extraction for LaTeX papers, improving discoverability and enabling richer metadata in analytics pipelines. Minor quality improvements include a deduplication refactor using dict.fromkeys and black formatting fixes. No major bugs fixed this month; emphasis was on feature delivery and overall code quality.