MarkSmith — The Markdown-to-Office Compiler That Never Corrupts

RECORDED FROM A RELEASE BUILD

This is the app. Actually running.

No mockup, no animation — captured off the real MarkSmith window in split view. The document streams in over the local POST /api/ingest endpoint, the same channel the browser extension uses. Then the Looking Glass opens, and a sentence typed through the aperture lands in all three places at once: the glass, the rendered page behind it, and the Markdown editor in the other pane.

MarkSmith in split view: a circular portal cut through the rendered page reveals the Markdown source, which also appears in the editor pane on the left — The document arrives section by section and the preview re-renders live — equations to native OMML, Mermaid to vector shapes. Then a portal is cut through the rendered page, and a sentence is typed straight into the source behind it: watch the word count on the editor pane tick up as the characters land. One edit, three surfaces, no mode switch.

Screenshots, straight off the window

Every image below is a PrintWindow capture of the running Release build — nothing composited, nothing hand-drawn. Click any shot to enlarge or download the real compiled DOCX files to test in Word.

Split view: Markdown source on the left, the rendered document preview on the right — **Split view & Looking Glass** Source and the Looking Glass preview side by side, rendering exactly what Word will produce. Download sample DOCX

MarkSmith identifying a pasted ChatGPT export and reporting the fixes it will apply — **It knows what pasted that** Drop in a raw ChatGPT export and the status bar names the source with a confidence score and a fix count — citation pips like `[12†source]`, stray `:contentReference` markers and `` math delimiters, all cleaned on the way to Word.
Download Without MarkSmith (Raw) Download With MarkSmith (Cleaned)

A Mermaid architecture diagram compiled into native editable Word shapes — **ShapeForge™ output in Word** A Mermaid diagram compiled to real DrawingML shapes. Select a node in Word and drag it — it is a shape, not a PNG. Download product-spec.docx

LaTeX matrices, integrals and chemistry equations rendered in the MarkSmith preview — **Math that survives the trip** Matrices, integrals and mhchem compiled to native OMML — editable in Word's equation editor, not pasted images. Download math-cheatsheet.docx

A MarkSmith-generated DOCX open in Microsoft Word with equations, code and a complex table — **The gauntlet document, in Word** Native equations, syntax-highlighted code and a nested table, opened in Word with zero repair prompts. Download gauntlet DOCX

Mermaid network architecture diagram rendered inside MarkSmith — **ShapeForge™ network solver** Mermaid network topologies compiled into vector geometry with automated connector routing and node fills. Download enterprise DOCX

MarkSmith house-style themed preview with corporate template styling — **Corporate house style engine** Live theme engine previewing branded fonts, paragraph spacing, and custom heading aesthetics before compiling. Download branded DOCX

Word-exact side-by-side rendering comparison — **Word-exact fidelity** What you preview is mathematically identical to what Word renders on disk. No surprise font shifts or table reflows.

MarkSmith AI prompt detection and citation cleanup tool — **AI prompt & citation cleaner** Sanitizes hallucinated citations, converts `` to standard LaTeX math, and strips leaked Markdown wrappers.

MarkSmith formatting engine and typography settings — **Typography & formatting inspector** Configure proportional column widths, heading numbering, code block shading, and WCAG AA contrast rules.

OpenXML schema governance and XML inspector — **Schema governance & XML inspector** Strict ECMA-376 schema enforcement. Dynamic `.rels` manager guarantees zero 'File is corrupted' prompts.

MarkSmith Express The same engine behind a local web UI — switch output profile between Word, HTML, slides, and eBook.

ARCHITECTURAL CAPABILITIES

Engineered for Zero Corruption & Maximum Fidelity

Word's rendering engine is notoriously fragile. MarkSmith was built from the ground up to strictly obey the ECMA-376 OpenXML schema with mathematical guarantees.

Mermaid network diagram compiled to native Word shapes

ShapeForge™ Vector Engine

ShapeForge™ Diagram Engine

Translates Mermaid flowcharts, mindmaps, state diagrams, and Gantt charts into native Word grouped vector shapes (<wpg:wgp>). Never a blurry PNG. In Word, click any node to drag it, recolor it, or edit text directly.

<wpg:wgp><wps:wsp><wps:spPr><a:prstGeom prst="roundRect"/>...

$LaTeX matrices and equations rendered in MarkSmith$

Native OMML Equations

Translates complex LaTeX math expressions directly into native Microsoft Office Math elements (<m:oMath>) with literal whitespace preservation (xml:space="preserve"). 100% editable inside Word's Equation Tools.

<m:oMath><m:r><m:t xml:space="preserve">\mathbf{M} = ...</m:t></m:r></m:oMath>

AI prompt detection and citation cleaner

AI Export Sanitizer

AI Prompt & Citation Sanitizer

Pasting raw ChatGPT or Claude exports breaks Word converters with unparsed [12†source] citation pips, leaked :contentReference tags, and raw  delimiters. MarkSmith normalizes the AST before Word ever sees it.

Regex Sanitizer: [12†source] & $ \theta $ → Cleaned Word AST

.dotx Master Injection

Corporate Template Injection

Injects compiled Markdown runs into existing corporate .dotx / .docx masters. Automatically inherits company fonts, heading hierarchies, cover pages, headers, and footers without manual formatting overhead.

<w:style w:type="paragraph" w:styleId="Heading1"><w:rPr><w:rFonts w:ascii="..."/>

ECMA-376 Governance

Zero-Corruption Guarantee

Word throws "The file is corrupted" errors whenever relationship IDs (rId) are hardcoded or namespaces are omitted. MarkSmith guarantees valid ECMA-376 schema generation with dynamic relationship management and $O(1)$ SAX streaming via OpenXmlWriter.

Namespaces: w:, a:, wp:, m:, wpg:, wps: | Dynamic rId in .rels

Formatting engine and typography inspector

KPI Cards & Geometry

Executive KPI Cards & Proportional Tables

Compiles :::metrics blocks into multi-column executive card grids with 24pt bold metrics, rounded shaded panels, and muted subtitles. Calculates dynamic proportional column widths across w:gridCol and w:tcW for rock-solid, zero-clipping table typography.

<w:tcW w:w="4800" w:type="dxa"/><w:gridCol w:w="4800"/> (Zero-Clipping)

CAPABILITY AUDIT · SEPTEMBER 2026

What Actually Separates These DOCX Engines

A side-by-side technical evaluation of five DOCX engines across document generation, shape compilation, OMML math, and template fidelity.

SIDE-BY-SIDE VISUAL PROOF

The Document Gauntlet: Pandoc vs. MarkSmith OpenXML Fidelity

Inspecting real Microsoft Word outputs compiled from the exact same Markdown document.

ISO/IEC 29500 Clean

Side-by-side comparison of Word outputs: Pandoc rasterized vs MarkSmith native vector shapes

Visual Diff Inspection • Click to enlarge high-resolution comparison Enlarge Diff

Key Takeaway: Pandoc embeds diagrams as dead raster PNG images. MarkSmith emits native Word DrawingML vector shapes (<wpg:wgp>) that remain fully editable.

Download Verified DOCX Source Markdown

Capability	MarkSmith	SuperDoc	Pandoc	HTML–to–DOCX	Word Interop
What it fundamentally is	Markdown → DOCX generator. Authors new documents from text.	DOCX editor & SDK. Round-trips documents that already exist.	Universal document converter. Multi-format conversion engine.	HTML serialiser. Converts web markup to document runs.	Word Application automation. Drives Microsoft Word via COM.
Microsoft Word required	No Fully offline, standalone compiler.	No Browser, Node, Python, CLI, MCP.	No Standalone binary.	No Library-level parser.	Yes Licence and runtime on every machine.
Diagrams from source text	Native vector Editable DrawingML shapes, not pictures.	— Out of scope Renders existing charts/shapes; no authoring.	Rasterised Filter output lands as PNG images.	None Not supported.	OLE objects Clunky embedded binary objects.
Math as native Office OMML	Authors OMML `LatexToOmml.cs` — LaTeX in, `<m:oMath>` out.	Reads OMML OMML→MathML converter for display.	Authors OMML Via `texmath`. Genuine parity here.	Raw text Unformatted text.	Partial Limited automation hooks.
Corporate template inheritance	.dotx, full Styles and headers inherited.	Field population Template fields, filling & locking.	.docx reference Carries styles, margins, headers, footers.	Strips styles Overwrites styles with raw HTML.	Manual macro glue Requires VBA macro glue.
SmartArt & .glox parsing	GLOX solver Built in — `GloxBuilder`, `OpenXmlDiagramGenerator`.	None No GLOX authoring path.	None Not supported.	None Not supported.	None Not supported.
Memory model, 1000+ pages	In-memory DOM High-speed AST tree in memory.	Bounded windows Claims progressive reads without browser DOM.	In-memory AST In-memory AST tree representation.	OOM at scale Crashes on high-volume runs.	Memory leaks COM handles leak under load.
WCAG contrast on output	AA 4.5:1 ContrastGuard, in the compose path.	Not on output Accessibility guide covers docs site, not DOCX.	None No contrast validation.	None None.	None None.
Edit an existing DOCX	No Generator, not an editor.	Core capability Writes back to OOXML with no HTML round-trip.	Convert only Lossy conversion only.	No Creation only.	Yes Full document manipulation.
Track changes & revisions	Native OpenXML CriticMarkup to <w:ins>, <w:del> & word/comments.xml.	Full Suggestions, comments, Yjs collaboration.	None Not supported.	None Not supported.	Via Word Native Word revisions model.
Frontier AI Dialect (Gemini 3.8 / Claude 3.7 / R1)	Native Word Folds Word-native collapsible <w15:collapsed> thought folding, search grounding footnotes & execution cards.	None Raw markup leak or unparsed XML tags.	None Stripped or unparsed HTML tags in DOCX.	None Drops or malforms AST elements.	None No AI dialect or reasoning awareness.
Executive KPI Cards & Proportional Tables	Native OpenXML Grid Native `:::metrics` 24pt executive dashboard cards & mathematical proportional column widths.	None Manual tables only; no metrics cards.	Auto-Fit Heuristics No metric cards; cell widths rely on Word layout engine.	None No metric cards; rough CSS table translation.	Programmatic Only No markdown syntax; requires manual cell coding.
Licensing model	Proprietary. Commercial license.	AGPLv3 / Commercial Dual-licensed SDK.	GPLv2+ Open-source binary.	Varies by tool MIT / Apache variants.	Commercial, per seat Microsoft Office license.

Native / verified

Partial or different mechanism

Absent

— Outside the product’s category

SuperDoc is not a converter, and that cuts both ways

It is a DOCX editor and agent SDK — the file already exists, and a person or an agent changes it. Six of the original eight rows were chosen where a generator naturally wins, so the table flatters MarkSmith by measuring SuperDoc at something it never set out to do. Saying that plainly is what makes the rows that do matter believable.

Where MarkSmith genuinely leads

Generating diagrams, SmartArt and math from plain source text, with contrast enforced on the way out. SuperDoc has no authoring path for any of it.

Where SuperDoc genuinely leads

Round-trip editing, track changes, comments and real-time collaboration — plus a Python SDK and an MCP server. MarkSmith has none of these.

Memory & Local Performance

MarkSmith compiles directly in-memory via .NET 8, eliminating COM overhead and generating full multi-page OpenXML documents in milliseconds.

How each capability was evaluated

MarkSmith capabilities were verified against the production MarkSmith.Core engine: ContrastGuard, GloxBuilder, LatexToOmml and the .dotx template path all run natively in .NET 8. SuperDoc capabilities derive from its technical documentation and repository specifications. Pandoc capabilities derive from its official manual and the texmath docx writer.

The Markdown-to-Office Compiler
That Never Corrupts.

This is the app. Actually running.

Screenshots, straight off the window

Test Drive the Engine. Verify Native Office Output.

Engineered for Zero Corruption & Maximum Fidelity

ShapeForge™ Diagram Engine

Native OMML Equations

AI Prompt & Citation Sanitizer

Corporate Template Injection

Zero-Corruption Guarantee

Executive KPI Cards & Proportional Tables

What Actually Separates These DOCX Engines

The Document Gauntlet: Pandoc vs. MarkSmith OpenXML Fidelity

SuperDoc is not a converter, and that cuts both ways

Where MarkSmith genuinely leads

Where SuperDoc genuinely leads

Memory & Local Performance

How each capability was evaluated

Choose Your MarkSmith Edition

Transform Your Markdown Documents Today

The Markdown-to-Office Compiler That Never Corrupts.

This is the app. Actually running.

Screenshots, straight off the window

Test Drive the Engine. Verify Native Office Output.

Engineered for Zero Corruption & Maximum Fidelity

ShapeForge™ Diagram Engine

Native OMML Equations

AI Prompt & Citation Sanitizer

Corporate Template Injection

Zero-Corruption Guarantee

Executive KPI Cards & Proportional Tables

What Actually Separates These DOCX Engines

The Document Gauntlet: Pandoc vs. MarkSmith OpenXML Fidelity

SuperDoc is not a converter, and that cuts both ways

Where MarkSmith genuinely leads

Where SuperDoc genuinely leads

Memory & Local Performance

How each capability was evaluated

Choose Your MarkSmith Edition

Transform Your Markdown Documents Today

The Markdown-to-Office Compiler
That Never Corrupts.