builder

tbdocs

Node.js static site generator for docs.twinbasic.com. Replaces the original Jekyll + just-the-docs pipeline (retired in the Phase 10 cutover; the legacy Ruby source set has since been removed from the tree).

Quickstart

Requires Node.js 22+. There is no builder/package.json; the whole dependency set lives in one package.json at the repository root.

npm ci                                    # from the repository root
node builder/tbdocs.mjs --src docs        # -> docs/_site/, _site-offline/, _site-pdf/

The day-to-day drivers are build.bat, serve.bat, check.bat and book.bat, all at the repository root -- they invoke tbdocs with the right arguments and route the browser-dependent gates and the PDF render behind the same wrappers. build.bat passes --check-audit-index, so an ordinary build also link-checks every tree it produced.

CLI flags:

Flag	Effect
--src <path>	Source root (default docs).
--dest <path>	Online tree destination (default <src>/_site). The offline tree lands at <dest>-offline, the PDF tree at <dest>-pdf, each removed instead when --no-offline or --no-pdf skips it. Only the default docs/_site rewrites the committed baselines and the add-in's index copy; another --dest compares and reports. It may not be or contain <src>, and inside <src> it must be, or be inside, a folder directly under it whose name starts with _site, _serve or _pdf; the build refuses any other.
--baseurl <prefix>	Overrides _config.yml's baseurl (used by CI to inject the GitHub Pages base path on fork deployments).
--url <origin>	Overrides _config.yml's url (used by CI so canonical URLs match the deployment origin).
--dry-run	Skip every filesystem write.
--no-offline	Skip the offline-tree pass (Phase 7).
--no-pdf	Skip the PDF-tree pass (Phase 8).
--tolerate-missing-images	Downgrade Phase 8's missing-image error to a warning.
--profile-offline	Per-substep timing for Phase 7.
--fetch-assets / --no-fetch-assets	Force remote-asset vendoring on or off. Default: download on a dev box, never when $CI is set.
--check / --no-check	Run (or skip) the link + integrity check over the HTML the build already holds in memory. Sets the exit code, never aborts the graph.
--check-audit-index	Implies --check; also diffs the derived tree index against what landed on disk. What build.bat passes.
--check-findings <path>	Implies --check; writes the findings as JSON for scripts/check_links_diff.mjs.
--serve	Start the long-lived dev server (watch + rebuild + SSE live-reload).
--port <N>	HTTP port for --serve mode (default 4000).

Serve mode

tbdocs --serve starts a long-lived dev process: HTTP server on port 4000 (override with --port), recursive watcher on the source tree, debounced rebuild on changes, and SSE-driven browser auto-reload. The offline and PDF passes are skipped each rebuild (restore them with a non---serve invocation).

node builder/tbdocs.mjs --src docs --serve

Or via the wrapper at the repository root: serve.bat.

Ctrl+C exits cleanly (closes the server, aborts the watcher, drains SSE clients).

Documentation

• PLAN.md -- architecture overview and the 12-phase pipeline.
• PLAN-1.md .. PLAN-13.md -- per-phase specs (inputs, outputs, edge cases, acceptance checklists). Phases 1-8 are the build itself; Phase 9 was the QoL consolidation pass; Phase 10 was the Jekyll cutover; Phase 11 is the output-changing parity update; Phase 12 adds --serve watch mode; Phase 13 is build-time SVG inlining with the zoom / export controls.
• PLAN-sab-pull-scheduler.md -- the shipped scheduler, and maintained reference rather than a frozen design record: pull model, SAB layout, Phases 5-18, and worked examples that are corrected when the code moves -- most recently to the landed registerBarrier form in tbdocs.mjs. It names pipeline tasks throughout; read them as current. PLAN-scheduler.md is the superseded push-based design, and PLAN-scheduler-offline.md covers moving the offline rewrite into the render workers.
• PLAN-checks.md -- folding the link checker into the build's task graph; Phase A shipped, the axe follow-ons are designed but not implemented.
• PLAN-a11y.md -- the WCAG 2.2 AA work. PLAN-axe-perf.md -- the axe-core performance investigation, complete.
• REVIEW-*.md / PLAN-REVIEW-*.md -- frozen audit snapshots of a commit range, not maintained reference documentation.
• FUTURE-WORK.md -- open follow-ups, grouped by divergence investigations / deferred enhancements.
• vendor/just-the-docs/README.md -- the vendored theme: what was taken, every in-tree patch to its _sass/ and just-the-docs.js with the commit that made it, and the re-vendoring procedure.

The end-user-facing documentation about how the build pipeline works lives on the site itself under Documentation Development: Tools and Scripts and its seven group pages are the cheat sheet; tbdocs Internals is the architecture overview; Extending the Builder is the how-to for adding a pipeline task, a markdown-it plugin or a render-worker sub-stage.

Build phases

Phase	Module(s)	Job
1	discover.mjs + publish-policy.mjs	Read .md/.html + frontmatter, enumerate static files, refuse any that is not a publishable type
2	nav.mjs / seo.mjs / book.mjs / build-info.mjs / data.mjs	Compute nav tree, SEO, book chapters, git commit info, _data/*.yml
3	render.mjs + highlight.mjs + highlight-theme.mjs	Markdown -> HTML body
4	template.mjs + compress.mjs	Wrap in layout, anchor-heading injection, whitespace compress
5	write.mjs	Write _site/
6	redirects.mjs / sitemap.mjs / search.mjs / symbols.mjs + symbol-baseline.mjs + addin-index.mjs	Auxiliaries (stubs, sitemap.xml, search-data.json, tB/symbols.json, the drift guard on its URLs and the help add-in's copy of it)
7	offline.mjs	Mirror to _site-offline/ with file:// URL rewrites
8	pdf.mjs + book.mjs (renderer half)	Sparse _site-pdf/ tree (book.html + CSS + images)

Two seed tasks run alongside discover: dot.mjs regenerates stale *.svg from their .dot siblings via the WASM build of Graphviz (with dot-metrics.mjs installing Inter's real advance widths first, so boxes are sized for the font the browser will paint), and vendor-assets.mjs downloads any referenced-but- uncommitted YouTube poster frame or GitHub user-attachment image -- never under CI, where a missing asset is a hard error instead.

Verification

Two gates abort the build outright, because their failure means the output itself is wrong rather than merely incomplete: nav.mjs's permalink and nav-integrity checks, and publish-policy.mjs, which holds the allowlist of file types that may reach a published tree. Everything under docs/ that is not a page is copied into the output verbatim, so a denylist only refuses what somebody named in advance -- before the allowlist, a stray .bak, .pem, .twin or frontmatter-less .md published at a public URL on a green build. It is enforced over the static-file inventory in discover and over each tree's derived inventory in dispatch, neither behind --check. scripts/check_publish_policy.mjs is the gate on that gate: a clean build says only that nothing is currently refused, which is also what an allowlist widened to nothing would say.

Regression detection runs inside the build. tbdocs --check (which build.bat passes) walks each tree's final HTML on the worker lane that produced it -- offline link check + duplicate-id detection, anchor resolution, remote-asset detection, sitemap, search-index and canonical integrity. A failing check sets the exit code but never aborts the graph. check.bat then runs the gates that need a browser. A clean build.bat && check.bat is the bar for "ready to commit".

test.bat is the other wrapper, and it is the one that judges an edit made here: it runs the gates that test the toolchain rather than the site, so a change under builder/ owes it whether or not any page moved. Two of its gates key on this directory by name. scripts/check_regex_safety.mjs refuses a regex that can backtrack exponentially --- both that shipped lived in render.mjs, and each hung a render worker in silence rather than failing, so the gate asks the question of the pattern instead of waiting for a page to ask it. It reads constructed patterns too, so assembling one from string constants is not an escape. scripts/check_code_regions.mjs refuses a pre-render rewrite that alters the contents of a code fence or code span, which four rewrites in render.mjs have done --- so a new one belongs inside applyPreRenderRewrites, between maskCode and its restore. Its probes also hold the rewrites over rendered HTML, which go through replaceOutsideCode from code-guard.mjs, to leaving a raw <pre> or <code> as written. Tools and Scripts lists every gate in both wrappers, in running order.

The pure core lives in link-check.mjs, and the check over it in check.mjs; check-tree.mjs is the build-side index. scripts/check_links.mjs runs the same check as a standalone tool, for trees this build did not produce. Neither CI workflow invokes it directly any more -- both reach it through check_links_diff.mjs, which runs it in-process against the fixtures. scripts/check_links_diff.mjs is the gate that says the two agree -- run it whenever any of the three changes; test/README.md covers the fixtures that give it something to disagree about, the exact finding counts they must keep provoking, and why a template change can turn the gate red without touching the fixture or the checker. See PLAN-checks.md.

The per-phase verify-phase{1..8}.mjs harnesses and the bulk-triage tools (_triage.mjs, _diff.mjs, etc.) were retired in the Phase 10 cutover; they asserted byte-equivalence with Jekyll, which is no longer the acceptance bar.

Every gate above checks the site. eval/ checks the documentation about the site: it hands an evaluator a goal and a mirror of the repository with all source stubbed unreadable, and asks whether the goal can be reached. That is a different question from whether the prose is accurate, and it catches a class of defect no gate here can see -- a page can be true in every sentence and still send a reader to the wrong remedy. See eval/README.md for how to run a round, and REVIEW-USECASES-874896e.md for what the first one found.