AI-Run Modern LaTeX Manuscript Workflow and Replication Packet

This record contains a public, reusable workflow note and sanitized replication material for an AI-run, human-directed pipeline that modernizes mathematical manuscripts into source-checkable LaTeX, reader PDFs, translations, audit packages, GitHub mirrors, and Zenodo releases. The packet documents the practical workflow: source acquisition and indexing, chunking, TeX transcription, compilation, visual/text auditing, translation passes, reader/artifact packaging, Zenodo publication, GitHub mirroring, and preservation of provenance artifacts. 2026-06-09 local-to-web audit update. The current scalable workflow uses local machines to prepare source-faithful aid packets for web-based Pro review threads. The local lane prepares source slices, page maps, rasterized context pages, 400-600 dpi crops for formulas/diagrams/tables, OCR prose witnesses, math-OCR candidates, TeX anchors, render checks, checksums, and status ledgers. The review lane uses those packets for translation, repair, and self-audit against the source. OCR witness rule. OCR, Pix2Text/Pix2TeX-style formula output, VLM output, detector candidates, and AI draft TeX are witness or locator layers until a source-specific audit promotes them. OCR is especially useful as a prose-gap witness: it can show that a draft skipped or compressed source text, but the promoted repair still needs source-image comparison and TeX compilation. High-resolution evidence rule. Diagram-heavy and formula-heavy pages often need high-detail evidence. A 150-200 dpi full-page image can provide page context, but small symbolic details, commutative diagrams, long arrows, tables, accents, subscripts, and superscripts often require labelled 400-600 dpi crops. The method is scalable only when the local lane sends targeted crops and ledgers rather than indiscriminate image dumps. Scan-quality rule. Bad scans or overcompressed images can create plausible but wrong outputs. Before repeating a repair/translation pass, check whether the source image is legible enough and whether a better scan or crop is needed. The 2026-06-09 Lean addendum remains part of the workflow record: Lean/Lake is a selective formal companion layer for stable theorem statements, definitions, examples, algebraic identities, and calculation lemmas after the TeX/source branch has already been source-compared. It is not a bulk proof that scanned pages, diagrams, typography, historical notation, or prose-level mathematical intent were transcribed faithfully. Corrections and improvements can be suggested through the GitHub repository via issues or pull requests: https://github.com/KokunoYumeto/modern-latex-manuscripts.