feat(markdown): strict mode tightens punctuation spacing

Strict mode already diverges from docling for readability (underscore
un-escaping, code-fence languages). Extend it to tighten stray spaces around
punctuation: `[ 37 , 36 ]` → `[37, 36]`, `( x )` → `(x)`, `*c* .` → `*c*.`.
This cleans up both the PDF backend's glyph-split citation spacing and the space
the legacy emphasis serialization leaves before punctuation.

Default/legacy output is untouched (still byte-for-byte with docling, which
keeps those spaces). Only inline text nodes are affected — code blocks and table
cells are left alone. Regenerated the strict regression fixtures; added a unit
test pinning legacy-spacing vs strict-tightened.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: artiz

crates/fleischwolf-core/src/markdown.rs

@@ -61,16 +61,25 @@ pub fn to_markdown_images(
     (md, ctx.artifacts)
 }
 
-/// In `strict` mode, undo the legacy `\_` underscore escaping the backends bake
-/// into inline text. Legacy output keeps `\_` (byte-for-byte with docling, which
-/// escapes underscores); strict prefers literal `_` for readability. Only inline
-/// text nodes are escaped — code blocks and table cells are left alone.
+/// In `strict` mode, rewrite inline text for readability rather than byte-for-byte
+/// docling fidelity: undo the legacy `\_` underscore escaping, and tighten stray
+/// spaces around punctuation (`[ 37 , 36 ]` → `[37, 36]`, `( x )` → `(x)`). This
+/// cleans up both the PDF backend's glyph-split spacing and the space the legacy
+/// emphasis serialization leaves before punctuation (`*a* ,` → `*a*,`).
+/// Legacy/default output keeps docling's spacing untouched. Only inline text
+/// nodes pass through here — code blocks and table cells are left alone.
 fn strict_text(text: &str, strict: bool) -> String {
-    if strict {
-        text.replace("\\_", "_")
-    } else {
-        text.to_string()
+    if !strict {
+        return text.to_string();
     }
+    text.replace("\\_", "_")
+        .replace(" ,", ",")
+        .replace(" .", ".")
+        .replace(" ;", ";")
+        .replace(" )", ")")
+        .replace("( ", "(")
+        .replace(" ]", "]")
+        .replace("[ ", "[")
 }
 
 fn render(nodes: &[Node], blocks: &mut Vec<String>, ctx: &mut Ctx) {
@@ -363,4 +372,14 @@ mod tests {
         // Strict prefers literal underscores (Rust-only readability mode).
         assert_eq!(doc.export_to_markdown_with(true), "# a_b\n\nx_y\n\n- i_j\n");
     }
+
+    #[test]
+    fn strict_tightens_punctuation_spacing_legacy_keeps_it() {
+        let mut doc = DoclingDocument::new("t");
+        doc.add_paragraph("see [ 37 , 36 ] and ( x ) .");
+        // Legacy keeps docling's spacing byte-for-byte.
+        assert_eq!(doc.export_to_markdown(), "see [ 37 , 36 ] and ( x ) .\n");
+        // Strict tightens punctuation for readable Markdown.
+        assert_eq!(doc.export_to_markdown_with(true), "see [37, 36] and (x).\n");
+    }
 }

crates/fleischwolf/tests/data/docx/expected/docx_checkboxes.docx.strict.md

@@ -14,4 +14,4 @@ Tasks before release:
 
 - [x] Implementation
 
-- [ ] Documentation of the task with a very long text that goes beyond the maximum length of a single line.
+- [] Documentation of the task with a very long text that goes beyond the maximum length of a single line.

crates/fleischwolf/tests/data/docx/expected/docx_rich_cells.docx.strict.md

@@ -15,7 +15,7 @@ Before table
 | Simple cell upper left     | Simple cell with  **bold**  and  *italic*  text      |
 | A B C Cell 1 Cell 2 Cell 3 | Rich cell A nested table  A B C Cell 1 Cell 2 Cell 3 |
 
-After table with **bold** , underline , ~~strikethrough~~ , and *italic* formatting
+After table with **bold**, underline, ~~strikethrough~~, and *italic* formatting
 
 ### Table with pictures
 

crates/fleischwolf/tests/data/docx/expected/equations.docx.strict.md

@@ -1,4 +1,4 @@
-This is a word document and this is an inline equation: $A= \pi r^{2}$ .
+This is a word document and this is an inline equation: $A= \pi r^{2}$.
 
 First item with inline equation: $A= \pi r^{2}$ is the area formula.
 
@@ -18,7 +18,7 @@ $$f\left(x\right)=a_{0}+\sum_{n=1}^{ \infty }\left(a_{n}\cos(\frac{n \pi x}{L})+
 
 This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text. This is text.
 
-This is a word document and this is an inline equation: $A= \pi r^{2}$ . If instead, I want an equation by line, I can do this:
+This is a word document and this is an inline equation: $A= \pi r^{2}$. If instead, I want an equation by line, I can do this:
 
 $$\left(x+a\right)^{n}=\sum_{k=0}^{n}\left(\genfrac{}{}{0pt}{}{n}{k}\right)x^{k}a^{n-k}$$
 
@@ -32,7 +32,7 @@ This is text. This is text. This is text. This is text. This is text. This is te
 
 This is a word document and these are inline equations: $N_{s}^{H}$ / $N_{s}^{P}$ ​. If instead, I want an equation by line, I can do this:
 
-$$e^{x}=1+\frac{x}{1!}+\frac{x^{2}}{2!}+\frac{x^{3}}{3!}+ \text{ \textellipsis } , - \infty <x< \infty$$
+$$e^{x}=1+\frac{x}{1!}+\frac{x^{2}}{2!}+\frac{x^{3}}{3!}+ \text{ \textellipsis }, - \infty <x< \infty$$
 
 And that is an equation by itself. Cheers!
 
@@ -60,7 +60,7 @@ $$\oiiint_{C}^{}\frac{1}{z}dz$$
 
 Operators used with limits:
 
-$$argmax_{ \epsilon }f(x),   \lim_{n}{\left(1+\frac{1}{n}\right)}^{n} ,  \max_{0 \leq x \leq 1}xe^{-x^{2}},  unsupported_{n}{\left(1+\frac{1}{n}\right)}^{n}$$
+$$argmax_{ \epsilon }f(x),   \lim_{n}{\left(1+\frac{1}{n}\right)}^{n},  \max_{0 \leq x \leq 1}xe^{-x^{2}},  unsupported_{n}{\left(1+\frac{1}{n}\right)}^{n}$$
 
 Equations with the OMML group character object:
 

crates/fleischwolf/tests/data/docx/expected/textbox.docx.strict.md

@@ -32,7 +32,7 @@ Upon receiving a report from the kindergarten, conduct a preliminary assessment
 Department of Education:
 Collaborate with the Health Bureau in conducting epidemiological investigations and assist Health Bureau personnel in implementing necessary epidemic prevention measures at all school levels.
 
-The Health Bureau will handle **reporting and specimen collection** .
+The Health Bureau will handle **reporting and specimen collection**.
 
 **Whether the epidemic has eased.**
 

crates/fleischwolf/tests/data/epub/expected/epub_purvis_poetry.epub.strict.md

@@ -1,6 +1,6 @@
 # Poetry
 
-By **Sarah Louisa Forten Purvis** .
+By **Sarah Louisa Forten Purvis**.
 
 <!-- image -->
 
@@ -10,13 +10,13 @@ The Standard Ebooks logo.
 
 <!-- image -->
 
-This ebook is the product of many hours of hard work by volunteers for [Standard Ebooks](https://standardebooks.org/) , and builds on the hard work of other literature lovers made possible by the public domain.
+This ebook is the product of many hours of hard work by volunteers for [Standard Ebooks](https://standardebooks.org/), and builds on the hard work of other literature lovers made possible by the public domain.
 
-This particular ebook is based on digital scans from the [Internet Archive](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry#page-scans) .
+This particular ebook is based on digital scans from the [Internet Archive](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry#page-scans).
 
-The source text and artwork in this ebook are believed to be in the United States public domain; that is, they are believed to be free of copyright restrictions in the United States. They may still be copyrighted in other countries, so users located outside of the United States must check their local laws before using this ebook. The creators of, and contributors to, this ebook dedicate their contributions to the worldwide public domain via the terms in the [CC0 1.0 Universal Public Domain Dedication](https://creativecommons.org/publicdomain/zero/1.0/) . For full license information, see the [Uncopyright](uncopyright.xhtml) at the end of this ebook.
+The source text and artwork in this ebook are believed to be in the United States public domain; that is, they are believed to be free of copyright restrictions in the United States. They may still be copyrighted in other countries, so users located outside of the United States must check their local laws before using this ebook. The creators of, and contributors to, this ebook dedicate their contributions to the worldwide public domain via the terms in the [CC0 1.0 Universal Public Domain Dedication](https://creativecommons.org/publicdomain/zero/1.0/). For full license information, see the [Uncopyright](uncopyright.xhtml) at the end of this ebook.
 
-Standard Ebooks is a volunteer-driven project that produces ebook editions of public domain literature using modern typography, technology, and editorial standards, and distributes them free of cost. You can download this and other ebooks carefully produced for true book lovers at [standardebooks.org](https://standardebooks.org/) .
+Standard Ebooks is a volunteer-driven project that produces ebook editions of public domain literature using modern typography, technology, and editorial standards, and distributes them free of cost. You can download this and other ebooks carefully produced for true book lovers at [standardebooks.org](https://standardebooks.org/).
 
 ## The Grave of the Slave
 
@@ -200,7 +200,7 @@ This bursting heart relief?
 
 There's nothing left for me to love
 This earth holds nothing dear,
-Since *he* , my sweet-my gentle one,
+Since *he*, my sweet-my gentle one,
 Is now no longer here.
 
 My poor fond heart had counted on
@@ -465,34 +465,34 @@ The Standard Ebooks logo.
 
 *Poetry*
 was compiled from poems published between 1831 and 1836 by
-[Sarah Louisa Forten Purvis](https://en.wikipedia.org/wiki/Sarah_Louisa_Forten_Purvis) .
+[Sarah Louisa Forten Purvis](https://en.wikipedia.org/wiki/Sarah_Louisa_Forten_Purvis).
 
 This ebook was transcribed and produced for
 [Standard Ebooks](https://standardebooks.org/)
 by
-[Weijia Cheng](https://weijiarhymeswith.asia/) ,
+[Weijia Cheng](https://weijiarhymeswith.asia/),
 and is based on digital scans from the
-[Internet Archive](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry#page-scans) .
+[Internet Archive](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry#page-scans).
 
 The cover page is adapted from
-*Sunday Morning* ,
+*Sunday Morning*,
 a painting completed in 1877 by
-[Thomas Waterman Wood](https://en.wikipedia.org/wiki/Thomas_Waterman_Wood) .
+[Thomas Waterman Wood](https://en.wikipedia.org/wiki/Thomas_Waterman_Wood).
 The cover and title pages feature the
 **League Spartan** and **Sorts Mill Goudy**
 typefaces created in 2014 and 2009 by
-[The League of Moveable Type](https://www.theleagueofmoveabletype.com/) .
+[The League of Moveable Type](https://www.theleagueofmoveabletype.com/).
 
 This edition was released on
 **May 21, 2026, 6:29** **p.m.**
 and is based on
-**revision 936122b** .
+**revision 936122b**.
 The first edition of this ebook was released on
 December 6, 2023, 8:52 p.m.
 You can check for updates to this ebook, view its revision history, or download it for different ereading systems at
-[standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry) .
+[standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry](https://standardebooks.org/ebooks/sarah-louisa-forten-purvis/poetry).
 
-The volunteer-driven Standard Ebooks project relies on readers like you to submit typos, corrections, and other improvements. Anyone can contribute at [standardebooks.org](https://standardebooks.org/) .
+The volunteer-driven Standard Ebooks project relies on readers like you to submit typos, corrections, and other improvements. Anyone can contribute at [standardebooks.org](https://standardebooks.org/).
 
 ## Uncopyright
 
@@ -504,4 +504,4 @@ Copyright pages exist to tell you that you *can't* do something. Unlike them, th
 
 Copyright laws are different all over the world, and the source text or artwork in this ebook may still be copyrighted in other countries. If you're not located in the United States, you must check your local laws before using this ebook. Standard Ebooks makes no representations regarding the copyright status of the source text or artwork in this ebook in any country other than the United States.
 
-Non-authorship activities performed on items that are in the public domain-so-called "sweat of the brow" work-don't create a new copyright. That means that nobody can claim a new copyright on an item that is in the public domain for, among other things, work like digitization, markup, or typography. Regardless, the contributors to this ebook release their contributions under the terms in the [CC0 1.0 Universal Public Domain Dedication](https://creativecommons.org/publicdomain/zero/1.0/) , thus dedicating to the worldwide public domain all of the work they've done on this ebook, including but not limited to metadata, the titlepage, imprint, colophon, this Uncopyright, and any changes or enhancements to, or markup on, the original text and artwork. This dedication doesn't change the copyright status of the source text or artwork. We make this dedication in the interest of enriching our global cultural heritage, to promote free and libre culture around the world, and to give back to the unrestricted culture that has given all of us so much.
+Non-authorship activities performed on items that are in the public domain-so-called "sweat of the brow" work-don't create a new copyright. That means that nobody can claim a new copyright on an item that is in the public domain for, among other things, work like digitization, markup, or typography. Regardless, the contributors to this ebook release their contributions under the terms in the [CC0 1.0 Universal Public Domain Dedication](https://creativecommons.org/publicdomain/zero/1.0/), thus dedicating to the worldwide public domain all of the work they've done on this ebook, including but not limited to metadata, the titlepage, imprint, colophon, this Uncopyright, and any changes or enhancements to, or markup on, the original text and artwork. This dedication doesn't change the copyright status of the source text or artwork. We make this dedication in the interest of enriching our global cultural heritage, to promote free and libre culture around the world, and to give back to the unrestricted culture that has given all of us so much.

crates/fleischwolf/tests/data/html/expected/formatting.html.strict.md

@@ -1,13 +1,13 @@
 # HTML Text Formatting Examples
 
-This is a **bold (b)** example and right next to it we have a **strong emphasis (strong)** . Notice that **strong + bold mixed** looks similar but carries additional semantic meaning.
+This is a **bold (b)** example and right next to it we have a **strong emphasis (strong)**. Notice that **strong + bold mixed** looks similar but carries additional semantic meaning.
 
-Here is an *italic (i)* word and an *emphasis (em)* example. Sometimes we combine them like *italic + emphasis together* .
+Here is an *italic (i)* word and an *emphasis (em)* example. Sometimes we combine them like *italic + emphasis together*.
 
-Now let's look at text that appears crossed out: ~~strikethrough with s~~ and ~~deleted with del~~ . You can also mix them: ~~double strikethrough (s + del)~~ .
+Now let's look at text that appears crossed out: ~~strikethrough with s~~ and ~~deleted with del~~. You can also mix them: ~~double strikethrough (s + del)~~.
 
-To highlight insertions or underlines: underlined with u , inserted with ins . A combination could be: underline + insertion together .
+To highlight insertions or underlines: underlined with u, inserted with ins. A combination could be: underline + insertion together.
 
-Subscript and superscript examples: Water is written as H 2 O using sub. The mathematical expression x 2 + y 3 uses sup. They can also be combined: CO 2 * .
+Subscript and superscript examples: Water is written as H 2 O using sub. The mathematical expression x 2 + y 3 uses sup. They can also be combined: CO 2 *.
 
-Mixing several: This sentence has ***strong + emphasis*** , some **bold + underline** , and a formula like a 2 + b 3 .
+Mixing several: This sentence has ***strong + emphasis***, some **bold + underline**, and a formula like a 2 + b 3.

crates/fleischwolf/tests/data/html/expected/html_code_snippets.html.strict.md

@@ -1,10 +1,10 @@
 # Code snippets
 
-The Pythagorean theorem can be written as an equation relating the lengths of the sides *a* , *b* and the hypotenuse *c* .
+The Pythagorean theorem can be written as an equation relating the lengths of the sides *a*, *b* and the hypotenuse *c*.
 
 To use Docling, simply install `docling` from your package manager, e.g. pip: `pip install docling`
 
-To convert individual documents with python, use `convert()` , for example:
+To convert individual documents with python, use `convert()`, for example:
 
 ```
 from docling.document_converter import DocumentConverter
@@ -20,5 +20,5 @@ The program will output: `## Docling Technical Report[...]`
 Prefetch the models:
 
 - Use the `docling-tools models download` utility:
-- Alternatively, models can be programmatically downloaded using `docling.utils.model_downloader.download_models()` .
+- Alternatively, models can be programmatically downloaded using `docling.utils.model_downloader.download_models()`.
 - Also, you can use download-hf-repo parameter to download arbitrary models from HuggingFace by specifying repo id: `$ docling-tools models download-hf-repo ds4sd/SmolDocling-256M-preview Downloading ds4sd/SmolDocling-256M-preview model from HuggingFace...`

crates/fleischwolf/tests/data/html/expected/hyperlink_01.html.strict.md

@@ -1,3 +1,3 @@
 # Something
 
-Please follow the link to: [This page](#) .
+Please follow the link to: [This page](#).

crates/fleischwolf/tests/data/html/expected/hyperlink_06.html.strict.md

@@ -1,7 +1,7 @@
 # Anchor Links Test
 
-Jump to [Section 2](#section-2) or visit [Example](https://example.com/) .
+Jump to [Section 2](#section-2) or visit [Example](https://example.com/).
 
 ## Section 2
 
-Content for section 2 with a [top link](#) .
+Content for section 2 with a [top link](#).