jss-lint: Automated Style Checking for Journal of Statistical Software Manuscripts

Manuscripts submitted to the Journal of Statistical Software (JSS) must satisfy a comprehensive style guide. This preprint presents jss-lint, an open-source, deterministic style checker for JSS-format manuscripts: a single Rust core implementing 62 rules in 16 categories for .tex, .Rnw, and .Rmd sources, shipped as a command-line tool, Python package, R package (CRAN: jsslintr), VS Code extension, GitHub Action, and a fully local WebAssembly browser application. The rules were written entirely by large language models under human direction and hardened by an evaluation-improvement loop against a pinned corpus of 254 real JSS-format manuscripts. The paper reports measured precision — 97.2% as a label-set estimate over 20,060 adjudicated violation instances, bracketed by the human-only subset (93.8%) and a labeler-error-propagated band (85.4%–91.2%) — and recall of 80.7% per instance against a hand-annotated ground-truth corpus, and documents the development methodology (specification-driven development, adversarially checked AI labeling, per-rule precision gates with documented exemptions and rule retirement) in detail. A replication script reproduces every number, table, and listing shown. jss-lint is an independent, third-party project; it is not affiliated with, endorsed by, or connected to the Journal of Statistical Software, its editors, or its publisher.