Self-Aware Writing

The problem

The reader's own attention is the subject, so the interface has to follow progress and memory without accounts or tracking — and the prose makes claims about itself that a later edit can quietly falsify.

What I owned

• Concept, writing and product design — the book and the software are one artifact
• Astro/TypeScript architecture where a chapter is one Markdown file everything else derives from
• Custom remark/rehype plugins for stanza breaks and in-chapter page splitting
• Reader behaviour: two reading modes, theme, progress, resume, offline PWA
• A checker for the claims the book makes about itself, plus 75 Playwright tests run across desktop, mobile and Safari

Technical challenges

• Chapters quote each other, so editing one file can falsify a sentence in another
• The text promises a moment with no words on screen, which the page break has to deliver
• Splitting pages at the AST level so the rail, pager and progress agree in both modes

Engineering decisions

• Astro's file-format build, so the rebuild kept the URLs the hand-written site had published
• Continuous reading by default and paged on request, with every page in the HTML either way
• Self-checks run in CI, so a contradiction between chapters fails the build
• Reader state in localStorage only — no analytics, no accounts, nothing leaving the browser

Technology