Manuscript Studio

Author-side book tooling, demonstrated on a sample chapter invented for this page. Every paragraph is content-addressed: its SHA-256 hash is registered on load, so any edit — even one character — is detected and must be explicitly accepted as a new revision. Sentences are machine-classified into typed claims feeding a verification queue, and a concept map is derived live from word co-occurrence. Edit the text below and watch all three react.

Chapter draft (editable — click into a paragraph)

Integrity registry

Claim verification queue

Concept map

What is running

The production version of this backs a book-length manuscript on Cloudflare Workers: paragraphs are stored content-addressed so the knowledge API serving excerpts can prove it is quoting the manuscript verbatim, edits create explicit revisions rather than silent drift, and factual sentences are extracted into a typed claim table (quantitative, attributed, normative) that a human works through before publication. The point is the discipline: prose you can hash is prose you can trust an automated system to serve.

Why this matters, in plain terms: in the age of AI, the hardest question about any text is "is this exactly what the author wrote, and can I check the claims in it?" This tooling answers both mechanically: a fingerprint on every paragraph proves nothing was silently altered, and every factual claim is pulled into a checklist a human verifies before the text is ever quoted by an automated assistant. It is the difference between a book an AI can responsibly answer questions about and a pile of words it might misquote.