AI-Run Modern LaTeX Manuscript Workflow and Replication Packet

This record contains a public, reusable workflow note and a sanitized replication packet for an AI-run, human-directed pipeline that modernizes public-domain mathematical manuscripts into source-checkable LaTeX, reader PDFs, translations, and archival release packages.

The packet documents the current practical workflow: scan acquisition and indexing, chunking, TeX transcription, compilation, visual/text auditing, translation passes, Zenodo publication, GitHub mirroring, preservation of provenance artifacts, and current OCR/math-OCR experiments. The 2026-05-31 refresh adds the concrete local document-tooling notes used for public workflow packets: Pandoc, LibreOffice/soffice, Python pdf2image, and Poppler-compatible PDF rendering checks.

The workflow is model-agnostic. It records the observed role of web LLM sessions, Codex-style local automation, Kimi-style large agent swarms, Claude Code-style local repair agents, Python/PyMuPDF tooling, TeX engines, GitHub, and the Zenodo API. OCR tools are described as witnesses only; mathematical fidelity still requires compilation, visual audit, and source checking against scans. The 2026-06-04 refresh adds reusable OCR/math-OCR scripts and lessons, including local GPU/OCR experiment notes and page-crop/formula-crop witness workflows.