FEBio-Documentation

A MkDocs site converting FEBio's LyX-authored manuals to Markdown, built to mirror the conventions of the sibling febio-feature-manual repository (Material for MkDocs theme, indigo palette, pymdownx.arithmatex + MathJax for equations, footnote-based citations). The site has two tabs, each a separate converted manual:

• Theory — the FEBio Theory Manual. Started as a single-chapter pilot (Chapter 2, Continuum Mechanics); now covers the complete manual — Chapters 1 through 8 plus Appendix A (Tensor Calculus).
• Studio — the FEBio Studio Manual. Started as a 2-chapter pilot (Introduction, Getting Started); now covers the complete manual — Chapters 1 through 20 plus Appendices A (Mesh Import Formats) and B (Standard Data Fields). See CONVERSION_NOTES_STUDIO.md for the full per-chapter breakdown and the real converter gaps this manual's content surfaced.

Both manuals are converted by the same generic, stdlib-only converter (tools/lyx2md.py), run once per manual by build.py with each manual's own source/output paths — see "How the converter works" below.

Table of Contents

• Repository layout
• Prerequisites
• Building the manual
• Deployment
• How the converter works
• Conversion statistics
• Known limitations / needs manual review
• Validation performed

Repository layout

source/ vendored copies of each manual's LyX source + BibTeX bibliography
FEBio_Theory_Manual.lyx Theory Manual (the complete manual; from febiosoftware/FEBio's Documentation/ dir)
FEBio3.bib
FEBioStudio_User_Manual.lyx Studio Manual (from febiosoftware/FEBioStudio's Documentation/ dir)
FEBioStudio.bib
tools/lyx2md.py the converter (stdlib-only, generic -- see "How the converter works")
build.py runs the converter once per manual (see its MANUALS list), generates mkdocs.yml
docs/ generated Markdown SOURCE for mkdocs -- this is mkdocs's input, not the
deployed site; see "Deployment" below
index.md site root landing page (not manual-specific; links to both tabs)
theory/index.md Theory Manual Preface (hand-authored)
theory/chapter<N>/*.md Theory Manual generated pages
studio/index.md Studio Manual Preface (hand-authored)
studio/chapter<N>/*.md Studio Manual generated pages
js/mathjax_config.js, febio.png -- the header logo, vendored from febio-feature-manual's docs/
tools/_stats.json Theory Manual conversion stats (see build.py's MANUALS list)
tools/_stats_studio.json Studio Manual conversion stats
.github/workflows/deploy.yml GitHub Actions workflow that builds and deploys to the gh-pages branch

Each manual's .lyx/.bib pair is vendored (checked-in) under source/, so this repository builds standalone from a bare git clone — the converter does not depend on any sibling directory outside this repo. (tools/lyx2md.py will also pick up the Theory Manual from a sibling ../febio-docs/ directory instead, if present and no --lyx/--bib flags are given, which is how it was used during local development against the original workspace layout — see the module docstring.)

Which chapters actually get converted into pages is controlled per-manual by the "chapters" entry in build.py's MANUALS list (passed through to tools/lyx2md.py --chapters), either a comma-separated list or "all". Both manuals now use "all": the Theory Manual's {1..9} (chapter 9 is the source's \start_of_appendix-marked chapter, rendered as "Appendix A") and the Studio Manual's {1..22} (chapters 21 and 22 are \start_of_appendix-marked, rendered as "Appendix A" and "Appendix B"). Chapters not in a manual's set are still scanned for their titles and label positions (so numbering and cross-references stay correct regardless of conversion order), they just don't produce output files yet — this is how both manuals grew from a small pilot to the full manual without ever needing to rewrite the cross-reference machinery. A chapter with no \begin_layout Section boundaries at all (Studio Manual's Appendix B is entirely Standard paragraphs and tables directly under the chapter heading) still produces a real output page — treated as a single synthetic section numbered <chapter>.1 — since mkdocs build --strict rejects an empty nav list outright rather than degrading gracefully.

Prerequisites

pip install mkdocs mkdocs-material

Building the manual

python3 build.py # runs tools/lyx2md.py, writes mkdocs.yml
mkdocs serve # preview at http://127.0.0.1:8000
mkdocs build --strict # build the static site into site/

build.py invokes tools/lyx2md.py as a subprocess once per manual in its MANUALS list, reads back each manual's own stats sidecar file (per-chapter, per-section formula/citation/figure counts and nav ordering), and uses that to generate a single mkdocs.yml with the correct navigation automatically — the nav never needs to be hand-maintained. Navigation uses navigation.tabs: each manual is exactly one top-level nav key (rendered as a tab — currently "Theory" and "Studio"), each containing a Preface entry followed by one nav group per converted chapter, expanding to that chapter's sections.

Deployment

The live site (https://wesley-jakob-gilbert.github.io/FEBio-Documentation/) is served by GitHub Pages from the gh-pages branch, not from the docs/ folder on main. docs/ on main is mkdocs's source input (Markdown); the gh-pages branch holds the fully rendered, compiled HTML output that mkdocs build produces into a local, gitignored site/ directory. These are two different branches with two different kinds of content — pushing to main alone does not, by itself, change what's live.

Two ways the gh-pages branch gets updated:

• Automatically:.github/workflows/deploy.yml runs on every push to main. It re-generates docs//mkdocs.yml from both manuals' vendored .lyx sources (so the deploy can never drift from what's actually committed), validates with mkdocs build --strict, then runs mkdocs gh-deploy --force, which builds the site and pushes the result to gh-pages. GitHub's own internal "pages build and deployment" step then republishes that branch to the live CDN — this second step happens outside our workflow and isn't always instant.
• Manually, e.g. to deploy local changes right away:

  mkdocs gh-deploy --force
  

  (requires push access to this repository).

How the converter works

tools/lyx2md.py is a stdlib-only, deterministic Python 3 parser for LyX's plain-text .lyx format. It does not shell out to LyX, Pandoc, or any other external tool. It's generic across manuals — CLI flags (--lyx, --bib, --docs-root, --nav-root, --stats-out, --chapters) select which manual's source it reads and where it writes, each defaulting to the Theory Manual's paths so a bare python3 tools/lyx2md.py is unchanged; build.py invokes it once per manual (see its MANUALS list) as a separate subprocess, so module-level globals and label registries never cross-contaminate between manuals. The numbered steps below describe the rendering logic itself, which is identical regardless of which manual is being converted.

1. Tokenize into a tree.parse_flat() reads the file line-by-line and builds an order-preserving tree of ('text', line), ('inset', spec, subitems), and ('layout', spec, subitems) tuples, tracking \begin_layout/\end_layout and \begin_inset/\end_inset nesting. Inline formula insets (\begin_inset Formula $x$ fully on one line) are special-cased at parse time so they don't get treated as the multi-line display-equation form.

2. Chapter/section boundary detection.main() locates every Chapter layout in the source (not just the ones being converted) to compute each one's absolute chapter number by position, then splits each converted chapter into its Section boundaries and numbers them <chapter>.<n>. Content that appears between the Chapter heading and the first Section boundary (chapter-level intro prose, and in one case — Chapter 6 — several numbered equations) is prepended to the first section's body rather than dropped; this was a real bug caught by reconciling formula counts chapter-by-chapter against the source (see CONVERSION_NOTES.md). A chapter whose source is marked with LyX's native \start_of_appendix layout (and every chapter after it) is numbered with a letter instead of a digit and labeled "Appendix" instead of "Chapter" in the nav — this is how Chapter 9 becomes "Appendix A" without hardcoding a chapter number.

3. Two-pass label resolution. Before rendering, a pre-scan pass walks every converted section and registers all \begin_inset CommandInset label targets (chapters, subsections, subsubsections, figures) into a global LABEL_REGISTRY, plus a parallel EQ_LABEL_REGISTRY for equation \label{}s (see the cross-section equation references bullet below), so \ref{}/\eqref{} cross-references — including ones that point to a different section's or chapter's file — can be resolved to the correct relative link, regardless of processing order. Every chapter lives in its own sibling directory under docs/theory/ (chapter1/, chapter2/, ...), so build_relative_link() computes same-page / same-chapter / cross-chapter links accordingly. A third registry, CHAPTER_LABEL_REGISTRY, is populated for every chapter in the source regardless of whether it's actually converted, so a \ref{} to a chapter (e.g. "see Chapter 5") always renders the correct chapter number — hyperlinked if that chapter has been converted, plain text if not — instead of the chapter's full title.

4. Character formatting (\series bold → **, \emph on → _, \shape italic → _, \family typewriter → `) is applied via a small state machine in render_items_inline(). LyX scopes this formatting to a single paragraph/layout and doesn't require an explicit closing toggle before \end_layout (confirmed: table cells routinely end with formatting left open), so any marker still open at the end of a call is auto-closed rather than leaking into whatever the caller appends next. A separate fix_emphasis_whitespace() post-pass hoists any whitespace LyX left inside emphasis/bold/code delimiters back outside them — Markdown (unlike LyX) requires no whitespace adjacent to the marker or it won't recognize the emphasis run at all. That pass masks underscores inside inline math ($t_{0}$-style LaTeX subscripts) before pairing markers, since otherwise they're indistinguishable from a real _..._ delimiter and can shift the pairing across an entire paragraph.

5. Math. Inline \begin_inset Formula $...$ becomes inline $...$; display insets (\begin{equation}, align, aligned, eqnarray, array) become \[ ... \] blocks with the LyX \label{} preserved inside, so MathJax's tags: 'ams' + \eqref{} numbering works exactly like in the Feature Manual. A same-page \eqref{}/\ref{} to an equation passes through as literal LaTeX for MathJax to resolve; a reference to an equation defined on a different page is resolved at build time instead (MathJax's per-page auto-numbering can't do this itself) — see the "Known limitations" bullet below. ref to a subsection becomes a Markdown link showing the subsection's title; ref to a figure becomes a Markdown link showing just the figure's number — its 1-indexed position among Graphics insets on its own page (matching what pymdownx.blocks.caption displays next to it), with a <section>. prefix only when the figure is defined on a different page than the reference, mirroring the equation-reference convention above. No "Figure" text is added, since the source prose always writes that word itself immediately before the \ref{} (e.g. "(Figure \ref{fig17})" or "Figure~\ref{...}a-c.").

6. Citations (\begin_inset CommandInset citation) become [^section-n] footnote references, deduplicated per page (the same BibTeX key cited twice on one page reuses one footnote number), with definitions resolved from that manual's .bib file via a small hand-written BibTeX field parser (parse_bib()), and appended at the bottom of each page as Author. "Title." *Journal* (Year).. \begin_inset CommandInset bibtex (LaTeX's \bibliography{} insertion marker, not an actual citation) is suppressed. Real footnotes (\begin_inset Foot, distinct from citations) are collected separately and appended as [^section-fn1], [^section-fn2], etc. Author/title/journal fields sometimes spell an accented letter as a raw LaTeX accent command rather than a literal Unicode character (e.g. {\"u} or \"u for u-umlaut, found in author names like "Gültekin"); decode_latex_accents() translates the common ones (umlaut, acute, grave, circumflex, tilde) to their real Unicode character, and residual {...} capitalization-protection braces around specific words/acronyms in a title (e.g. {A {Nonparametric} Approach}) are stripped, since no case-transformation is applied to citation text here that they'd need to protect against.

7. Figures (\begin_inset Float figure + \begin_inset Graphics) become ![name](figs/name.png) followed by a pymdownx.blocks.caption-style /// figure-caption block, with the figure's own \label hoisted to an <a id="..."> anchor placed before the image (rather than left inline in the caption prose, which is where LyX actually stores it). A LyX scale NN attribute is carried through as inline CSS ({: style="width:NN%" } via attr_list) so the figure isn't embedded at full native pixel size. A bare Graphics inset not wrapped in a Float (decorated instead with LyX Box/VSpace insets, which are otherwise purely presentational and rendered as their content passed through transparently) is also handled, as is \begin_inset Wrap figure (LaTeX's wrapfig text-wrapped figure, found in the Studio Manual) — Markdown has no text-wrap-around-image equivalent, so it renders identically to an ordinary Float figure, and the figure-numbering prescan counts it the same way so \ref{} numbering stays consistent with what actually appears on the page.

8. Tables (\begin_inset Tabular) become plain Markdown tables (first row as header), always wrapped in a centering <div markdown="1" style="display: flex; justify-content: center;"> — a plain Markdown table has no native alignment syntax, and attr_list doesn't attach to a table at all (confirmed empirically: an appended {: ... } gets absorbed as a bogus extra table row instead), so md_in_html is required in mkdocs.yml to get the nested table syntax inside that wrapper div actually parsed rather than passed through as literal text. LyX's tabular format is an embedded pseudo-XML dialect (<lyxtabular>, <row>, <cell>) that parse_flat() doesn't parse structurally; render_tabular() uses those tags purely as delimiters to group the Text insets (which are ordinary, correctly-parsed insets) holding each cell's real content, protecting row-separator newlines with a sentinel character so they survive a later prose-whitespace-normalization pass that would otherwise collapse them onto one line. Merged cells (colspan/rowspan) aren't representable in plain Markdown and are flagged for manual review rather than silently producing a misaligned table (occurs in the element-property tables of Section 4.1). A table wrapped in \begin_inset Float table (found throughout the Studio Manual; the Theory Manual only ever uses bare Tabular insets) is rendered the same way, via the same render_tabular() call, with its caption using pymdownx.blocks.caption's separate table-caption type (own numbering sequence, doesn't perturb figure \ref{} numbering).

9. ERT ("evil red text", raw LaTeX LyX has no native inset for) is reconstructed from its per-line \backslash-token encoding and handles the two patterns that occur in this document: \href{url}{text} (rendering a real Markdown link, unwrapping a nested \emph{} in the link text to Markdown emphasis) and a bare \url{url} (rendering as a Markdown autolink <url>). Anything else is flagged for manual review instead of guessed at. LyX's native hyperlink inset (\begin_inset CommandInset href, distinct from the ERT reconstruction above and used throughout the Studio Manual) follows the same convention: an optional display name becomes a Markdown link, an unnamed one becomes a bare autolink.

10. Unhandled inset kinds render as <!-- UNHANDLED INSET ... --> HTML comments and are logged to needs_review — this makes the required zero-leftover-artifact grep double as a completeness check: any real parser gap shows up as a grep-able marker instead of silently dropping content.

11. Headings get explicit anchor IDs via attr_list syntax (## Title {: #label }), which is why attr_list was added to mkdocs.yml's markdown_extensions beyond the Feature Manual's baseline set — it's required for the cross-section \ref{} links in step 3 to have a target to land on.

12. Theorem-style layouts. LyX's theorems-ams module layouts Example (numbered, per-page counter) and Theorem* (unnumbered) render as a bold run-in label directly in the text flow — **Example N.** <body> / **Theorem.** <body> — matching the published manual's plain LaTeX theorem-style numbering, not a Material admonition callout box (which the original document doesn't use; an earlier version of this converter rendered these as !!! example "Example N" boxes, since corrected). Per LyX/LaTeX semantics, consecutive same-kind layouts (nothing but a blank line between them) are additional paragraphs of the same environment instance, not a new one each — render_section_body() tracks the previous top-level layout's kind and only advances the counter / starts a fresh bold label when it wasn't the same kind (a non-blank item in between, e.g. LyX's \begin_deeper, still breaks the run, since that does mark a genuinely separate instance — confirmed against Appendix A.1, which has both cases). Paragraph layouts (an unnumbered run-in sub-heading, one level below Subsubsection) render as a bold #### heading. FormulaMacro insets (LyX Math Macro definitions, e.g. Chapter 7's 23 local shorthand macros) render as nothing — they're definitions, not visible content; the equivalent MathJax macros entries live in docs/js/mathjax_config.js instead, since MathJax has no per-page macro scoping.

13. Description layouts (LaTeX's description list environment, used throughout the Studio Manual; never occurs in the Theory Manual) render as a paragraph whose label (everything up to the first plain space) is auto-bolded, matching LyX's own rendering — unless the body already starts with **, since some source items explicitly wrap their label in \series bold instead of relying on auto-bolding, and auto-bolding again would double it up.

14. LyX-Code layouts (literal code/data listings — XML session-file snippets, CSV data rows, also Studio-Manual-only) render via render_code_line(), which skips the prose-oriented character-formatting state machine and whitespace normalization used everywhere else (both would corrupt significant indentation/whitespace in a literal listing). Consecutive LyX-Code layouts are grouped into a single fenced ``` block rather than one block per line, mirroring the Example/Theorem* continuation rule above.

Output: one Markdown file per converted Section, in docs/theory/chapter<N>/ for the Theory Manual (e.g. 2.1-vectors-and-tensors.md) or docs/studio/chapter<N>/ for the Studio Manual.

Conversion statistics

Theory Manual

Totals for the complete, fully-converted manual; see CONVERSION_NOTES.md for the full per-section breakdown.

Metric	Count
Chapters converted	9 (Chapters 1–8 plus Appendix A / Tensor Calculus)
Sections converted	64
Inline $...$ formulas emitted	5203
Display \[...\] formulas emitted	1919
Citations	203
Figures	21 (artwork fetched at build time — see below)
Unhandled/unknown inset kinds	0
Leftover LyX bookkeeping artifacts in output	0

See CONVERSION_NOTES.md for the full per-section breakdown and every item flagged for manual review.

Studio Manual

Totals for the complete, fully-converted manual; see CONVERSION_NOTES_STUDIO.md for the full per-chapter breakdown.

Metric	Count
Chapters converted	22 (Chapters 1–20 plus Appendices A and B)
Sections converted	117
Inline $...$ formulas emitted	126
Display \[...\] formulas emitted	18
Citations	2
Figures	194 (artwork fetched at build time — see below)
Unhandled/unknown inset kinds	0
Leftover LyX bookkeeping artifacts in output	0

See CONVERSION_NOTES_STUDIO.md for the full per-chapter breakdown and every real converter gap this manual's content surfaced (not present in the Theory Manual's source).

Known limitations / needs manual review

• Every chapter of the Studio Manual is now converted. It started as a 2-chapter pilot so the converter's support for this manual's LyX constructs (e.g. Wrap-figure insets, native CommandInset href links, Description/LyX-Code layouts, a Float table content-loss bug) could be validated against real content before committing to a full 20-chapter/117-section conversion — see CONVERSION_NOTES_STUDIO.md for the complete list of real converter gaps this manual's content surfaced, several of which also turned out to be latent bugs in the Theory Manual's own output (e.g. malformed bold/italic nesting when LyX doesn't close formatting markers in reverse-of-open order, \size <name> commands leaking as literal text, undecoded LaTeX accent escapes in citation author names).
• Figure artwork is fetched automatically at build time, per manual. Figures aren't part of either manual's original LyX/BibTeX inputs, so build.py scans every converted chapter's generated Markdown for figure references and fetches any missing ones from that manual's own upstream Documentation/Figures/ directory — febiosoftware/FEBio for the Theory Manual, febiosoftware/FEBioStudio for the Studio Manual — into that chapter's figs/ directory (skipping the fetch if a real copy is already present). The original LyX-authored captions are preserved intact either way.
• Cross-section (and cross-chapter) \eqref{}/\ref{} references to equations are resolved to static links, not left as \eqref{}. Each Section is a separately-loaded page, and MathJax's tags: 'ams' auto-numbering is per-page -- it has no way to resolve a reference to a \label{} defined on a different page, which renders as a bare "???" with nothing to click. EQ_LABEL_REGISTRY in tools/lyx2md.py tracks every labeled equation's 1-indexed position among its own page's AMS-numbered equations (verified to exactly match what MathJax itself displays, and cross-checked via a real browser that the link both navigates to the right page and lands on the right equation); such a reference is then replaced with a real link like (2.5-35) to the target page's MathJax-generated #mjx-eqn:<label> anchor, mirroring how the published manual itself handles the identical problem at its finer per-subsection pagination (there it reads (2.5.4-2)). MathJax sanitizes spaces in the label to underscores when building that anchor id (confirmed against the one label in this document that contains a space, eq:virtual work → mjx-eqn:eq:virtual_work) — mathjax_eqn_id() replicates that. Because the anchor is injected by MathJax after the browser's initial page-load fragment-scroll already ran (and thus failed to find it), docs/js/mathjax_config.js also re-attempts the scroll once typesetting finishes, via MathJax's startup.pageReady hook. Same-page references are untouched since MathJax already resolves those correctly on its own.
• Merged cells (colspan/rowspan) aren't representable in plain Markdown tables. The element-property tables in Section 4.1 use them; render_tabular() flags each occurrence for manual review and renders a best-effort approximation rather than silently producing a misaligned table.
• Every chapter is now converted, so cross-references no longer point outside the site. (Earlier notes in this file mis-described three references — eq87, eq:viscous-stress, eq:virtual work — as broken references in FEBio's own source; they were always real, resolvable labels, just in chapters not yet converted at the time. All three, and every other cross-reference in the manual, now resolve.) mkdocs build --strict still emits INFO-level "does not contain an anchor" messages for cross-page equation links — this is a static-checker false positive, not a broken link: #mjx-eqn:<label> anchors are injected client-side by MathJax only after it typesets the target page, so mkdocs's link checker (which only inspects the built HTML/Markdown source) can't see them, even though they resolve correctly in a real browser. Zero WARNING-level messages.
• \obslash has no LaTeX macro definition anywhere in the LyX source. The document's preamble defines \tr, \dev, \grad, \divg, etc. as \newcommands (and docs/js/mathjax_config.js reproduces them as MathJax macros so they render instead of leaving raw command names on the page), but \obslash — used 8 times in Chapter 2 for a tensor "conjugate" transpose-product operator — is never defined, even in the full manual. It was initially approximated as an overlined \oslash; visual comparison against the published manual showed the actual glyph is a backslash-in-a-circle rather than an overlined forward-slash-in-a-circle, so it now renders as U+29B8 CIRCLED REVERSE SOLIDUS (⦸), the mirror image of \oslash's U+2298 CIRCLED DIVISION SLASH. This depends on the MathJax web font covering that codepoint — reconfirm visually if the MathJax CDN version ever changes.
• \mbox{...} is aliased to keep its argument in math mode rather than switching to true text mode, because the LyX source nests math macros (\dot{}, \thinspace) inside \mbox{} in a few places, and MathJax's \text{} does not expand macros in its argument. It's aliased to \mathrm{#1} rather than a no-op group, so plain-word arguments (and, grad, div, M) render in upright text font while nested macros like \dot{} and \thinspace still expand correctly.
• mkdocs build --strict succeeds (exit code 0). The only diagnostics emitted are the INFO-level anchor messages described above — there are zero WARNING-level messages.

Validation performed

1. python3 build.py && mkdocs build --strict — exit code 0.
2. grep -rn '\begin_\|\end_inset\|\begin_inset\|SpecialChar\|\lang ' docs/theory/ — zero matches across all converted pages.
3. Zero unhandled/unknown inset kinds logged across all converted chapters.
4. Formula counts reported by the converter's own render-pass counters (not a post-hoc regex scan) were checked chapter-by-chapter against the source. Chapter 2 alone reconciles exactly (1455 inline + 368 display = 1823, matching grep -c "begin_inset Formula" over its source range exactly, as in the original single-chapter pilot); Chapter 6 also reconciles exactly (400/400) after the chapter-intro content-loss fix described above. Chapter 3 has one long-standing, tiny (1 of 1096, ~0.1%) unreconciled formula not further pursued — see CONVERSION_NOTES.md. Chapter 7's apparent 23-formula "gap" is fully explained by its 23 FormulaMacro definitions, which correctly produce no visible output.
5. A real browser (headless Chromium via Playwright) was used throughout development to verify things a static grep can't catch: MathJax equation/table/figure rendering, a chapter title containing inline math ($\alpha-$Method) rendering correctly in both the page heading and the nav sidebar, and that a cross-chapter equation reference link both navigates to the right page and scrolls to the right equation. (navigation.tabs was originally removed in favor of a single unified sidebar while this was a single-manual site; it was reintroduced when the Studio Manual pilot was added, this time with each manual as one top-level nav key/tab rather than one tab per chapter — see item 8 below.)
6. Sections 2.1 and 2.6 were read in full and spot-checked against the published manual at help.febio.org TM40-Section-2.1 and the Chapter 2 table of contents — see CONVERSION_NOTES.md for the detailed comparison. Every equation, definition, and the section 2.6 subsection ordering (2.6.1 through 2.6.9) match the published manual exactly.
7. The site was served locally with mkdocs serve and screenshotted with a headless Chromium (Playwright) during the original single-chapter pilot — see screenshot_section_2.1.png and screenshot_section_2.6.png in the repo root. Equations render as properly typeset math (fractions, matrices, tensor operators, numbered equations with working anchors) with no raw LaTeX source visible on the page.
8. Studio Manual pilot (Chapters 1–2) and navigation.tabs addition:python3 build.py && mkdocs build --strict — exit code 0, zero WARNING-level messages, run against both manuals together. The leftover-artifact grep from item 2 was re-run over all of docs/ (both docs/theory/ and docs/studio/) — zero matches. The built HTML (site/index.html) was checked for md-tabs__link elements confirming exactly two tabs render ("Theory", "Studio"). Two real converter gaps were found and fixed against this manual's actual content (not present in the Theory Manual's source, so never previously exercised): a Wrap-figure inset (LaTeX's wrapfig) had no renderer and was falling through to UNHANDLED INSET, now aliased to the existing Float figure renderer; and a whitespace-normalization regex intended to strip a spurious space before sentence-ending punctuation was also stripping the genuine space before a literal file-extension token ("the .xplt file extension" was rendering as "the.xplt..."), fixed with a negative lookahead so it only strips a space before punctuation not immediately followed by a word character.
9. Studio Manual widened from the 2-chapter pilot to the complete manual (all 20 chapters plus Appendices A and B, 117 sections):python3 build.py && mkdocs build --strict — exit code 0, zero WARNING-level messages. The leftover-artifact grep — zero matches across all 117 Studio Manual sections plus all 64 Theory Manual sections. Every section's needs_review entries were scanned and confirmed to be only the two routine, expected categories (figure-fetch placeholders, and — before every chapter was converted — forward references into not-yet-converted chapters); zero unexpected entries remained after the converter fixes below. A broad spot-check read roughly 15 sections spanning materials, contact, post-processing, mesh data, Python scripting, and both appendices for rendering quality beyond what the automated checks catch. Ten further real converter gaps were found and fixed this pass — see CONVERSION_NOTES_STUDIO.md for the complete list — the most significant being a content-loss bug where a table wrapped in a \begin_inset Float table (as opposed to the already-handled Float figure) was silently dropped entirely (caption and anchor still rendered, but the table itself vanished with no needs_review flag), and a formatting-nesting bug — also present in some pre-existing Theory Manual output — where LyX source that closes character-formatting markers out of LIFO order (e.g. closing \series bold while \emph on, opened after it, is still active) produced invalid Markdown nesting; fixed by replacing four independent boolean flags with an explicit open-marker stack.