PR #1430 — EPUB Joins the Book's Default Output Set

The Broader Goal at the End of the Previous PR

By PR #1409, "The Supervised Machine" had a real visual identity layered onto the HTML source of truth, and the PDF rebuild step was an established, documented ritual in docs/book/README.md. The book's distribution story was "HTML source, PDF render" — two formats, one manual Chrome headless-print command, committed together whenever the manuscript changed.

Strategic Shifts in This PR

None — this is a small, deliberate gap-close, not a repositioning. The book has been growing (multi-author chapters, Part II/III/IV expansions, embedded UI mockups) and PDF-only distribution doesn't serve e-readers well: no reflow, no adjustable type size, no proper e-reader TOC navigation. EPUB was the obvious missing format, and the user asked for it to become a default step, not an occasional extra — matching the existing "regenerate the PDF and commit both files together" discipline this project already holds for the HTML/PDF pair.

What This PR Shipped

• Installed pandoc via Homebrew (not previously present in this environment).
• Generated docs/book/the-supervised-machine-v0.5-DRAFT.epub from the existing single-file HTML source, with --resource-path=. (run from docs/book/) so the three rxcc Stitch mockup PNGs under assets/ embed correctly instead of producing PandocResourceNotFound warnings.
• The key technical catch: pandoc's EPUB writer splits into chapter files at <h1> by default, but this manuscript's real chapter headings are <h2 class="chap-title"> — the manuscript wraps everything in a handful of top-level <h1>s (front matter, Part dividers) with actual chapters one level down. Splitting at the default level produced a single 216KB chapter file with no real per-chapter navigation. Passing --split-level=2 fixed this, producing 36 properly split chapter files and a correct nav.xhtml TOC (verified by unzipping the EPUB and spot-checking that TOC entries map to the right chapter titles).
• Verified EPUB structural validity by hand (no epubcheck available in this environment): confirmed mimetype, META-INF/container.xml, EPUB/content.opf, EPUB/toc.ncx, and EPUB/nav.xhtml are all present and well-formed via unzip -l / unzip -p.
• Documented the exact rebuild command in docs/book/README.md, right alongside the existing PDF rebuild command, and updated the file list and "regenerate together, never let them drift" language to cover all three formats (HTML, PDF, EPUB) instead of two.

Brainstorm Visuals Used

None — this PR is build tooling, not new manuscript content.

What This Achieved on the Path to Autonomy

Nothing autonomy-related directly, but it closes a real distribution gap for a project asset that synlynk itself produces and maintains (the book is written by the same Claude/Codex/Grok/Agy fleet that builds synlynk). It also sets a precedent worth generalizing later: any generated-artifact pair (source + rendered output) in this repo should default to every commonly consumed target format, not just the first one someone thought to add.

Strategic Note: The Goal at the End of This PR

The book's default output set is now HTML (source of truth) + PDF + EPUB, generated and committed together on every content change. The next book-adjacent PR should treat EPUB regeneration as non-optional the same way PDF regeneration already is — no separate reminder needed, it's just part of "the book changed."