A scanned book is a stack of pictures. A search engine, an agent or a reader cannot cite a picture, and a plain text dump of it hides every misread word, lost formula and broken table.
Doclab Arcana reads a book and hands back a checked transcription: Markdown text, an anchor for every page, the real images, and a quality report that lists exactly which pages are damaged and how. It does not pretend a bad page is good — a book with problems is published marked as damaged, with the list of pages to fix.
It is the first step of the Knowledge Contract chain: digitise a book, convert it to clean formats, package it with its images and relations, and only then index it and put it into contracts.
- Books published
- 18
- Output
- Markdown, page anchors, images
- Runs on
- One Apple Silicon Mac
- Part of
- Knowledge Contract
01
Every page can be cited
The text keeps an anchor for each page and the original images, so a quotation can point to the exact page it came from.
02
Honest about damage
Each book carries a quality report. One book is clean; seventeen are published marked as damaged, with a per-page list of what still needs fixing.
03
Ready for the contract
A command exports a single book as a Knowledge Contract package.
Why a scan is not yet knowledge
Old and new books alike arrive as page images or as files whose text layer is unreliable. Optical recognition makes mistakes in exactly the places that matter: formulas, tables, footnotes, chemical notation.
Doclab Arcana treats those mistakes as data to be tracked, not noise to be hidden. Every book leaves with a record of what was read well and what was not.
A local stand, on purpose
Recognition runs on a local vision model and classic recognition tools, on a single Apple Silicon machine. Books never have to leave it to be read. Development happens elsewhere through reviewed changes, and the machine runs only the released version.
The price of that choice is plain: the stand is one computer, not a service.
What comes next
The next work, in order: reading formulas and chemistry in a chemistry textbook, re-reading whole tables, recovering words lost in footnotes, delivering Knowledge Contract packages for every published book, and a translation stage into Russian.
Capabilities
Working now
- Transcription of scanned books into Markdown with page anchors and real images
- A quality report for every book with a per-page list of damage
- Publishing finished bundles to a private books library
- Export of a single book as a Knowledge Contract package
Planned
- Transcription of born-digital books proven on library books
- Reading of formulas and chemical notation
- Whole-table re-reading for complex tables
- Recovery of words lost in footnotes
- Knowledge Contract packages delivered for all published books
- A translation stage into Russian
- Running as a service rather than on a single machine
Current autonomy level
Weakest link
The stand runs on one computer and only that computer. Seventeen of the eighteen published books are still marked damaged, and formulas, complex tables and footnotes are not read reliably yet. No autonomy level is assessed.
Roadmap to —
- Read formulas and chemistry correctly, starting with a chemistry textbook.
- Re-read whole tables instead of cell by cell.
- Recover words lost in footnotes.
- Deliver Knowledge Contract packages for all published books.
- Add a translation stage into Russian.
Next step
Review the current build and its development plan
BUILDING means active development, not a promise of public availability. Use the current capabilities and roadmap above as the reference.
Review the path to the target level