Skip to content
Olivier CarrèreTechnical writing practice

Site history

This history begins with the registration of redaction-technique.org in February 2012 and follows the documented publishing systems, content changes and architectural decisions that led to the current documentation portfolio.

The key milestones give the story in brief. The timeline holds the detailed record, drawn from the Git repository and from the content itself. Each statement says how it is known, and what the repository cannot tell is marked as such.

History covered
February 2012 – today
Recorded in Git
October 2014 – today
Publishing systems
3, plus 3 experiments
Content phases
10
Releases
7
Changes listed
48

Key milestones

The story in brief. Each milestone links to its detailed entry in the timeline. These notes place the site’s history in the wider evolution of technical writing and documentation.

  1. 2012

    1. Origin

      Domain registeredredaction-technique.org

    Asciidoctor begins

    Asciidoctor, a Ruby implementation of AsciiDoc, grows out of a prototype written for the Git website and has its first release in June. It helps make lightweight text sources the input of a programmable documentation toolchain.

    Sources: Asciidoctor historyAsciidoctor documentationAsciidoctor releases on RubyGems

  2. 2013

    1. Publishing system

      WordPress blog

    Docs-as-code takes shape

    The first Write the Docs conference, in Portland in April, marks the emergence of a community around documentation. The period sees the practices now called docs-as-code take shape: Git, plain-text markup, code review and automated checks applied to documentation.

    Sources: Write the Docs conferencesWrite the Docs origin storyWrite the Docs: Docs as Code

  3. 2014

    1. Publishing system

      reStructuredText / Sphinx site

    2. Experiment

      DITA XML version

    3. Experiment

      DocBook version, from the DITA branch

    Structured documentation matures

    DITA, an OASIS standard since 2005, gives professional documentation a model for structured, reusable content. DITA Open Toolkit 2.0 is released in November, a month after HTML5 becomes a W3C Recommendation, as the Web matures as a delivery platform for technical content.

    Sources: OASIS DITA Technical CommitteeDITA Open Toolkit releasesW3C HTML5 Recommendation

  4. 2021

    1. Event

      Last Sphinx state

    Docs-as-code becomes established

    By 2021, docs-as-code is an established practice in software documentation. Technical writers increasingly work in Git repositories, with pull-request review, automated checks and static-site generators, publishing documentation the way software is released.

    Sources: Write the Docs: Docs as CodeAnne Gentle: Docs Like CodeGitLab documentation workflow

  5. 2025

    1. Experiment

      First Astro attempt, on a side branch

    2. Publishing system

      Astro / Starlight documentation sitev2.0.0

    3. Event

      Documentation moves to docs.redaction-technique.org

    AI-assisted documentation accelerates

    AI use in documentation work moves from experiment towards practice, increasingly for drafting, summarising, search and review. Documentation is also published for AI agents to read, through llms.txt files and Model Context Protocol servers such as Microsoft Learn’s, launched in June 2025.

    Sources: Write the Docs Portland 2025 talksStack Overflow 2025 survey: AIThe llms.txt proposalMicrosoft Learn MCP Server

  6. 2026

    1. Release

      Information architecture and documentation platformv2.1.0

    2. Release

      Technical writing portfolio repositioningv2.2.0

    Documentation and software engineering converge

    Documentation increasingly overlaps with software engineering: kept in repositories, checked in continuous integration, and written for AI coding agents as well as people, as AGENTS.md files are. Talks at Write the Docs 2026 examine how documentation and large language models interact.

    Sources: AGENTS.mdLinux Foundation: Agentic AI FoundationWrite the Docs Portland 2026 talks

  7. Today

    1. Current state

How the site evolved

Newest at the bottom. Rows follow event order, not a time scale: long quiet stretches take one row.

How to read this history

Each publishing system opens with a heading, and each year with its number. Events are on the left. What changed in the content is on the right, or below on small screens.

  • Diamond: the origin of the history, the domain registration. It is not in Git.
  • Production line: the history the site is published from.
  • Large filled dot: a publishing system begins on the production line.
  • Not in Git: known only from indirect evidence.
  • No recorded Git activity on the production line. This does not mean the site was offline.
  • Side branch: an experiment that was never merged back.
  • Release: a tagged version of the whole site.
  • Continues today.

Commit references, evidence and uncertainties open from each entry’s disclosure, by click, tap or keyboard. How each statement is known: observed in Git or in the content, reported in the author’s words, inferred, or not established.

How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Interpretation. A reading of the evidence, not a direct observation.Uncertain. Not established by the evidence.

  1. Origin

    Domain registered

    redaction-technique.org

    Nothing is recorded between the registration and the earliest trace of the WordPress blog, in February 2013.

    How this is known: Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
    Evidence and uncertainties
    • Domain registration, recorded by hand by the site's author. It predates the repository: there is no commit for it. (Author’s words)

    Not established

    • What the domain served between the registration and the earliest trace of the WordPress blog, in February 2013.
    • The licence added to the imported files claims copyright from 2011, a year before the registration. The repository does not explain the difference.

  2. Publishing system 1 of 3≤ 2013?

    WordPress

  3. ≤ 2013?

    WordPress blog

    • Not in Git
    • Historical context

    The original site. Its source is not in the repository: what survives is the content imported on 2014-10-21, in French, as a single reStructuredText file of 106 sections.

    Git activity
    none: not in Git
    In production
    dates not recorded; still live on October 28, 2014
    How this is known: Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Uncertain. Not established by the evidence.
    Evidence and uncertainties

    Sphinx work began in Git on 2014-10-21 while WordPress was still live on 2014-10-28: the two overlap. When Sphinx replaced WordPress is not recorded.

    • "Add content from current www.redaction-technique.org website": a single 5,832-line redaction-technique-org.rst. (Git)
    • A WordPress media-library path in the imported text: the media library contains a file uploaded in February 2013. (Git)
    • Tag note, written 2014-10-28: "Still some glitches, but already better than current WordPress version." WordPress was still the live site then. (Author’s words)
    • The site states it "a été initialement développé sous WordPress" and that the CMS's lack of cross-cutting changes and lifecycle tracking led to the move to reStructuredText. (Author’s words)
    • The licence header added to every imported file claims copyright from 2011 ("Copyright 2011-2014"). (Author’s words)
    • The pruning commit removed "formerly unpublished sections", which suggests the import included unpublished material. (Author’s words, Interpretation)

    Not established

    • Start date. The licence claims copyright from 2011 and the oldest upload dates from February 2013. Neither proves when the WordPress site opened.
    • End date. WordPress was still live on 2014-10-28. When the Sphinx build replaced it is not recorded.
    • Post dates, categories, tags, authors and URLs did not survive: the import is ordered alphabetically by title and carries no per-article metadata.
    • Whether the "documentation-code-source-" filename prefix reproduces WordPress slugs or was added during the import.

    Content≤ 2013? – October 2014

    A French blog about documentation as source code

    The import contains 106 sections: technical writing as an industrial process, structured formats and DITA XML, conrefs and conditional text, FrameMaker to DITA migration, version control and group work, DITA Open Toolkit and XSL-FO tutorials, and marketing deliverables (white paper, brochure, company magazine, website).

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Interpretation. A reading of the evidence, not a direct observation.
    1 change in this phase

    1. Imported

      All content imported into one reStructuredText file, ordered alphabetically by title.

      commit 1fc63738Git

  4. Publishing system 2 of 32014–2021

    reStructuredText / Sphinx

  5. 2014

  6. reStructuredText / Sphinx site
    • Production line

    The imported content rebuilt as a single-sourced Sphinx project publishing HTML, PDF, and EPUB, and from 2015 a print edition. Versioned 0.1 to 1.5 in 2014-2015, then extended as a blog. The last Sphinx change is dated 2021-07-17.

    Git activity
    October 2014 – July 2021
    In production
    dates not recorded
    Hosting
    no hosting record in Git; evidence of a Read the Docs build (2014) and of a live Sphinx page (2025)
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Uncertain. Not established by the evidence.
    Evidence and uncertainties

    Neither the date the Sphinx build went live nor the date it was replaced is recorded. The repository holds evidence of a Read the Docs build in December 2014 and, in May 2025, a link to a live page of the Sphinx site, but no independent hosting record. The last Sphinx commit is 2021-07-17. The Astro conversion is 2025-09-13.

    • Split the imported file into one .rst file per section (106 files). (Git)
    • Tag note: "All text and code reviewed". (Author’s words)
    • PDF cover for the lulu.com print version. (Git)
    • Applied sphinx_rtd_theme. (Git)
    • The welcome page offers "une version compilée sur Read the Docs", linked to readthedocs.org/projects/redaction-techniqueorg. (Git, Author’s words)
    • The link is removed. Commit message: "Inline conditional text processing implemented in Makefile is not supported on Read the docs." (Author’s words)
    • Last reStructuredText commit. Tag note: "Latest Python Sphinx version". (Git)
    • The reStructuredText-Sphinx branch ends at this commit, which is also part of master's history: the branch marks the last Sphinx state rather than a separate line of work. (Git)
    • On the patch-1 side branch, the first Astro attempt's French About page links to the "version HTML" of the earlier content at redaction-technique.org/diminuer-les-couts-ameliorer-la-satisfaction-client.html, a page of the Sphinx site. A link is not a hosting record. (Git, Author’s words)

    Not established

    • Hosting. The welcome page linked to a compiled version on Read the Docs from 2014-12-19 to 2014-12-29, and a May 2025 link points to a live page of the Sphinx site, but Git holds no hosting record. The dates the Sphinx build went live and was replaced are not recorded.
    • How long the site was served from the Sphinx build after 2021-07-17. The May 2025 link suggests a Sphinx page was still online then. It does not show until when.

    Content

    Pruned and restructured as a three-part manual

    The imported blog became a three-part manual: reducing costs and improving customer satisfaction, technical writing as an industrial process, and the DITA XML structured format. The October-November 2014 clean-up dropped 38 imported sections, including the marketing-deliverable pages and near-duplicates. Files carried review-status comments, still present on 86 files in the last Sphinx state. Release 1.0 marked all text and code as reviewed.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Interpretation. A reading of the evidence, not a direct observation.
    5 changes in this phase

    1. Restructured

      Split into one file per section.

      commit 59f6bbd1Git

    2. Content deleted

      36 sections deleted. The commit calls them unpublished or no longer in line with the new site.

      commit e6ad350cGit, Author’s words

    3. Content deleted

      Deleted "Faciliter le travail de groupe avec Subversion, cron et Emacs".

      commit 7fa3d89dGit

    4. Content deleted

      Deleted "Tests de non-régression".

      commit bda8e681Git

    5. Reviewed

      Release 1.0: all text and code marked as reviewed (69 .rst files).

      commit ca619c6bGit, Author’s words

  7. Release

    0.1: tagged source on GitHubFirst version of the Sphinx site
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  8. WordPress still live

    • Not in Git

    The 0.1 tag note calls WordPress the "current" version.

    How this is known: Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.

  9. Release

    1.0: tagged source on GitHubAll text and code reviewed
    How this is known: Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.

  10. Release

    1.1: tagged source on GitHubStarting point of the DITA branch
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

    DITA XML version
    • Experiment
    • Parallel branch

    DITA XML branch starts from release 1.1.

    A parallel conversion of reStructuredText release 1.1 to DITA XML, set up to build a PDF with DITA Open Toolkit. The Sphinx line continued independently from the same release. The DITA branch was never merged back. A DocBook branch started from it on 2014-11-30 and stopped on 2014-12-09.

    Git activity
    November – December 2014
    Publication
    not known
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.Uncertain. Not established by the evidence.
    Evidence and uncertainties

    Whether the DITA version was ever published online is not recorded. Its README only describes a local PDF build.

    • Both the DITA branch and the continuing Sphinx line start from release 1.1. (Git)
    • README: "DITA XML version based on reST version 1.1". It describes building full.ditamap into a PDF with DITA-OT 1.8.5. (Git, Author’s words)
    • "Migrate to DITA XML": 50 topics with no code, plus no-code-topics.ditamap. (Git)
    • Removed the reStructuredText files on the DITA branch only. The Sphinx line kept them. (Git)
    • Hierarchical full.ditamap. Tasks added 2014-11-18 to 2014-11-20. (Git)
    • The only DITA release tag. (Git)
    • A DocBook branch starts from dita-0.1 ("Add DocBook version"). 7f55e388 then removes the DITA files from it. (Git)
    • Last commit on the DITA branch. (Git)

    Not established

    • Whether the DITA version was ever published online. The README only describes a local PDF build.
    • Editorial passes made on the branch (bold to italics, non-breaking spaces) also appear on the Sphinx line in December 2014 (67ba3281, 4cda9d57). Which one influenced the other cannot be proven.

    Content

    The same content, converted to DITA XML

    A side branch from release 1.1, running alongside the Sphinx line. Content converted, not extended: at the median, 89% of each DITA topic's wording is found in the matching .rst of release 1.1. Sections were typed as topics and tasks and assembled with ditamaps. No new article was written.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.
    2 changes in this phase

    1. Format conversion

      Topics without code converted to DITA.

      commit ad6c24edGit

    2. Format conversion

      Remaining topics added.

      commit db5a521bGit

    Content

    Editorial conventions and multi-format publishing

    Starting from release 1.1, the same release the DITA branch forked from: terminology moved into substitutions, external links were centralised, typography was normalised, the welcome page was rewritten, and graphics were rebuilt from SVG sources for HTML, PDF, EPUB, and print. Releases 1.1 to 1.5.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.
    3 changes in this phase

    1. Revised

      Welcome page rewritten and restructured.

      commit aff876afGit

    2. Content added

      Print edition for lulu.com.

      commit f77e40e6Git

    3. Restructured

      PNG and PDF images generated from SVG sources.

      commit 1d621175Git

  11. DocBook branch starts from dita-0.1

    • Experiment
    Git activity
    November – December 2014
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  12. The DITA and DocBook branches stop

    Neither was merged back into the production line. The Sphinx site went on.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  13. 2015

  14. Release

    1.5: tagged source on GitHubLast numbered Sphinx release
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  15. 291 days

    No recorded Git activity on the production line. This does not mean the site was offline.

  16. 2016

  17. Content

    From manual to blog: tutorials and an end-to-end project

    The first new articles and sections since the import: a "Le coin du geek" section (Python, Raspberry Pi), conditional-text and data-insertion tutorials using Jinja, SQL, Excel, and DITA, and a "Mener un projet de bout en bout" case study split into 14 pages, then 15. The site calls itself a blog from 2016.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
    9 changes in this phase

    1. Content added

      String manipulation with Python.

      commit 7951e910Git

    2. Content added

      Raspberry Pi as a documentation platform.

      commit 69c5b05eGit

    3. Content added

      New top-level section "Le coin du geek".

      commit 583f22a6Git

    4. Renamed

      Dropped the "documentation-code-source-" prefix from every file name.

      commit 7a867bc3Git

    5. Content added

      Start of the 2018 conditional-text and data-insertion tutorials.

      commit 68fb0652Git

    6. Site feature added

      Homepage box listing recent Git changes. An early form of today's change history.

      commit 92adcbf3Git, Interpretation

    7. Content added

      New section "Mener un projet de bout en bout".

      commit 951d843cGit

    8. Restructured

      The project section split into 14 pages.

      commit 69afa81eGit

    9. Content added

      A 15th project page, on Git.

      commit 16e63892Git

  18. 382 days

    No recorded Git activity on the production line. This does not mean the site was offline.

  19. 2017

  20. Git activity resumes on the production line

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  21. 272 days

    No recorded Git activity on the production line. This does not mean the site was offline.

  22. 2018

  23. Git activity resumes on the production line

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  24. 913 days

    No recorded Git activity on the production line. This does not mean the site was offline.

  25. 2021

  26. Content

    Last Sphinx articles

    Homepage articles on documentation as microservices and on personas. Neither carried over to Astro. The Sphinx site ends with 110 pages.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.
    2 changes in this phase

    1. Content added

      Documentation/microservices comparison.

      commit fa5c0a5eGit

    2. Content added

      Persona article on the homepage.

      commit 68956980Git

  27. Last Sphinx state

    The Sphinx history is the first part of master. The reStructuredText-Sphinx branch only marks its last state: it is not a separate product.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  28. 1519 days

    No recorded Git activity on the production line. This does not mean the site was offline.

    Outside the production line: patch-1 (May – July 2025). An unmerged branch started from the last Sphinx commit. It holds the first Astro attempt, shown below. Its later commits, from June 2025, add unrelated content. None of them reached master.

  29. 2025

  30. First Astro attempt, on a side branch

    • Experiment
    • Parallel branch

    Work on Astro and Starlight began in May 2025 on a side branch, patch-1, away from the production line. Pandoc converted 109 of the 110 Sphinx pages to Markdown. Two days later a small bilingual site was started from scratch instead, and the converted pages and the Sphinx sources were then removed from the branch. This version did not reach the production line: in September 2025 the conversion was done again from the last Sphinx state, and that conversion became the current site.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Interpretation. A reading of the evidence, not a direct observation.
    Evidence and uncertainties
    • "Convert ReStructuredText to Markdown", the branch's first commit: 109 Markdown files, one for each page of the last Sphinx state except "documentation microservices.rst". (Git)
    • "Rebuild blog from scratch". (Git)
    • The attempt's page "Migrate content from Python Sphinx to Astro": "I used Pandoc to convert it to Markdown automatically", then "I just decided to git rm everything and start fresh", as the content "didn't reflect state-of-the-art technical writing practices". (Author’s words)
    • Merge of pull request #1. This merge commit is on patch-1 only. Master does not contain it. (Git)
    • "Remove Sphinx content": the converted Markdown files and the .rst sources leave the branch. (Git)

    Not established

    • When the attempt was set aside.
    • Whether this version was ever published online.
    • Pull request #1 was merged on GitHub on 2025-05-11, but its merge commit is not on master. Whether it targeted another branch, or master was later reset, is not recorded.

  31. Publishing system 3 of 32025–today

    Astro / Starlight

  32. 2025

  33. Conversion to Markdown begins

    The last Sphinx state is converted to Markdown again, starting with one commit, and this conversion becomes the production site. It continues over the following days.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Content comparison. Measured by comparing the wording of pages across versions.Uncertain. Not established by the evidence.
    Evidence and uncertainties

    The Astro site continues the Sphinx production line on master. It was converted from the last Sphinx state, not from the DITA branch.

    Git records evidence of parts of the conversion toolchain, not a reproducible record of every step. The May 2025 attempt’s own page says pandoc converted the pages. 98 of the 109 Markdown files first produced in September match that output once line wrapping is ignored. And an unmerged branch keeps the later processing scripts.

    • The May 2025 attempt's own page: "I used Pandoc to convert it to Markdown automatically". (Author’s words)
    • "Update markdown files" on the unmerged starlight branch: 109 Markdown files, one per Sphinx page, before any Starlight processing. (Git)
    • convert2starlight.py, on the same branch: asks an OpenAI model to turn each Markdown file into a Starlight page. (Git)
    • "Create dir structure from Sphinx TOC", after toc.js and create_docs.sh were added. (Git)

    Not established

    • Every conversion step: the commands run between these commits, and any edits made by hand, are not recorded.

    Content

    Converted to Markdown and consolidated

    The Sphinx site's 110 pages became 63 French Markdown files in five sections, then 49 once duplicates were deleted on 2025-09-15. Many Sphinx sub-pages ended up inside their parent pages rather than in pages of their own: see the lineage rows labelled "folded into another page". An English version was drafted with AI and its images translated.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
    6 changes in this phase

    1. Format conversion

      Conversion from reStructuredText to Markdown begins.

      v2.0.0commit 67b3e056Git

    2. Content deleted

      Deleted 15 end-to-end project pages as duplicates: their text was already in the project's main page.

      v2.1.0commit 222a4c57Git, Author’s words, Content comparison

    3. Translated

      First draft English version.

      v2.1.0commit 291fd568Git

    4. Content deleted

      Deleted the standalone formats-cibles page: its text was already on the About page.

      v2.1.0commit 2237502aGit, Content comparison

    5. Translated

      AI-generated English version (pull request #3).

      v2.1.0commit 907ade6bGit, Author’s words

    6. Translated

      Images translated.

      v2.1.0commit aa3b1e09Git

  34. Release

    v2.0.0: tagged source on GitHubAstro + Starlight migration
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  35. Documentation moves to docs.redaction-technique.org

    From 16 September 2025, the documentation links to itself at docs.redaction-technique.org and points readers to a blog at www.redaction-technique.org. The blog is a separate project: its history starts the same day and shares no commit with this one.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Interpretation. A reading of the evidence, not a direct observation.
    Evidence and uncertainties
    • "change link to docs.redaction-technique.org": the homepage links point to docs.redaction-technique.org instead of redaction-technique.org. (Git)
    • The homepages link to a "Blog" at https://www.redaction-technique.org/, "A blog about all things documentation as code." (Git, Author’s words)
    • The blog's "Initial commit", on branches main and draft of the same repository: a history with no commit in common with master. Later that day it redirects its own links to docs.redaction-technique.org (b48b765c). (Git)
    • The documentation's configuration sets its address to https://docs.redaction-technique.org/. (Git)
    • The blog's configuration sets its address to https://redaction-technique.org. (Git)

    Not established

    • When each address began serving each site: DNS and hosting changes are not in Git.

  36. 254 days

    No recorded Git activity on the production line. This does not mean the site was offline.

  37. 2026

  38. Content

    English slugs, search metadata and editorial polish

    Pages in both languages got English slugs with redirects. Search metadata was improved, "Technology watch" became "Tutorials", prose in both languages was revised to the new style guide, and several tutorials were modernised with live in-browser demos.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.
    7 changes in this phase

    1. Restructured

      Search metadata improved across all English and French pages.

      v2.1.0commit 447b8ae8Git

    2. Renamed

      English slugs translated from French, with 301 redirects.

      v2.1.0commit a6e00789Git

    3. Content added

      Writing style guide.

      v2.1.0commit 3d49d457Git

    4. Renamed

      French slugs aligned with English ones.

      v2.1.0commit 8ae964dbGit

    5. Renamed

      "Technology watch" section renamed "Tutorials".

      v2.1.0commit 5c58c6b0Git

    6. Revised

      DITA XML/XSL-FO tutorials modernised for DITA-OT 4.4.

      v2.1.0commit 4775bf8bGit

    7. Content deleted

      Word-shuffling Python page deleted.

      v2.1.0commit 16b7d6e0Git

  39. Content

    Goal-based information architecture and documentation platform

    Navigation was reorganised around reader goals. Every article was given a concept, task, or reference structure. No prose was rewritten. New content: a practical toolkit and a glossary. New site features: an "Ask the documentation" assistant and a static documentation API. Diagrams redrawn, first in Mermaid, then as build-time SVG.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
    6 changes in this phase

    1. Site feature added

      "Ask the documentation" assistant.

      v2.1.0commit 8de98035Git

    2. Site feature added

      Static documentation API (llms.txt, JSON index).

      v2.1.0commit e9d7b09aGit

    3. Restructured

      Navigation by reader goal.

      v2.1.0commit 081b58beGit

    4. Restructured

      Articles given a concept/task/reference structure. According to the commit, no prose was rewritten.

      v2.1.0commit bc644c60Git, Author’s words

    5. Content added

      Toolkit, glossary and API documentation page.

      v2.1.0commit 2e4a841aGit

    6. Restructured

      All diagrams as native SVG components.

      v2.2.0commit 7c647c39Git

  40. Release

    v2.1.0: tagged source on GitHubInformation architecture and documentation platform
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

  41. Release

    v2.2.0: tagged source on GitHubTechnical writing portfolio repositioning
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

    Content

    Content audit and portfolio repositioning

    A 2026 content audit modernised the process pages in both languages, written separately for each language. The site was repositioned as a senior technical writer's portfolio, and browsing by topic, page change histories, and release milestones were added.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
    7 changes in this phase

    1. Revised

      English/French content audit: process pages modernised, legacy material given context.

      v2.2.0commit f45b0179Git

    2. Content deleted

      Obsolete "index in a PDF" page deleted.

      v2.2.0commit e742ee1dGit

    3. Reverted

      The archiving of seven legacy pages (9512f38d) was reverted the same day. The pages remain.

      v2.2.0commit fc99cdc3Git

    4. Content added

      Expertise page and portfolio homepage.

      v2.2.0commit 23c92671Git

    5. Site feature added

      Change history on documentation pages.

      Unreleasedcommit ee0a661fGit

    6. Site feature added

      Browse the documentation by topic.

      Unreleasedcommit d920814eGit

    7. Site feature added

      Git release milestones.

      Unreleasedcommit 33d3b631Git

  42. Today

    Astro / Starlight documentation site
    • Production line
    • Ongoing

    The current site. After a first attempt on a side branch in May 2025, the last Sphinx state was converted to Markdown in September 2025, starting with one commit. The conversion continued in the following commits, and master carried on from there. The site became bilingual (French and English) in September 2025, was restructured around reader goals in 2026, and repositioned as a technical writing portfolio.

    3 changes since v2.2.0 are not yet in a release.

    Git activity
    September 2025 – ongoing
    In production
    current; start not recorded
    First Astro work
    May 2025, on a side branch
    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Author’s words. The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.Uncertain. Not established by the evidence.
    Evidence and uncertainties

    The current site, deployed on Vercel. The date it replaced the Sphinx site in production is not recorded.

    • "Migrate to Àstro", made on the last Sphinx commit (68956980). Deletes 110 .rst files and adds a Starlight project (dimensional-disk/) holding the pages as Markdown, plus convert2starlight.py. (Git)
    • Merge of pull request #2. Release "Astro + Starlight migration". (Git)
    • Moved the Astro project to the repository root. (Git)
    • Vercel Analytics. (Git)
    • Release "Information architecture and documentation platform". (Git)
    • Release "Technical writing portfolio repositioning". (Git)

    Not established

    • The complete conversion toolchain. Git records evidence of parts of it (see the conversion), not a reproducible record of every step.

Current state

What the history above has led to.

  • Built with Astro and Starlight from Markdown and MDX sources kept in Git, and deployed on Vercel.
  • 74 pages in English and 74 in French, organised around reader goals.
  • Latest release: v2.2.0, on September 19, 2026.
  • Every push to master is built by a continuous-integration check, and an automated test suite checks the content, the documentation API and this history against Git.
  • Each documentation page shows its change history from Git, and this page is reconstructed from Git by a script that fails on any difference.
  • Diagrams are reusable Astro components drawn as SVG at build time. An “Ask the documentation” assistant and a static documentation API (llms.txt, JSON index) are built from the same sources.

What this history tells you

What the timeline shows about the content: what each change of publishing system kept, what has lasted longest, and how the diagrams changed.

What happened to the content?

Pages are tracked by their wording, not their file name: a page counts as carried over when enough of its text is found in a later version. The three stages happened years apart.

  1. 2014 clean-up

    The content imported from WordPress, before the Sphinx site’s first releases.

    106 sections imported from WordPress

    • 68kept
    • 38dropped in 2014 clean-upImported from WordPress and deleted in the October–November 2014 clean-up, before release 1.0. It never reached the later Sphinx site.

  2. Sphinx years, 2014–2021

    What the Sphinx site held in its last state.

    110 pages in the final Sphinx state

    • 68inherited from the WordPress import
    • 42written during the Sphinx years

  3. 2025 conversion and after

    What became of those 110 pages when the site moved to Markdown and Astro, and since.

    110 pages in the final Sphinx state

    at the 2025 conversion
    • 39kept as its own pageKept a page of its own through the 2025 conversion, and still has one.
    • 62folded into another pageHas no page of its own after the conversion. At least 30% of its text is found inside another page.
    • 6not carried overLess than 30% of its text is found anywhere in the converted or current site.
    after the conversion
    • 1kept, later rewrittenKept a page of its own in 2025, but today’s page retains less than half of the text the conversion carried over.
    • 2carried over, later deletedConverted in 2025, then deleted.

    Not part of this conversion: the 38 sections marked “dropped in 2014 clean-up” never reached the Sphinx site.

    At the 2025 conversion, the median page had last received an individual edit about 9.3 years earlier, and 34 of the 110 pages were last individually edited in 2014. Site-wide changes, such as copyright-year updates, do not count as individual edits. Much of what the conversion carried was long-lived documentation, not recent writing.

    How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Interpretation. A reading of the evidence, not a direct observation.

  4. Today

    74 French pages

    • 43carry text from the historical site
    • 31carry no text from the Sphinx site

    74 English pages

    67 of the 68 imported sections kept in 2014 are still in today’s site, as a page of their own or inside another page.

What survived

Release 1.0, in November 2014, organised the site in three parts. Their introductions still open three sections today:

The subjects endured, not the form: these pages were converted, restructured and revised. Around them, the tutorials once gathered in “Le coin du geek” now sit under Tutorials, the end-to-end case study is folded into a single page, and the diagrams, page addresses, navigation and homepage were all redone.

How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.Content comparison. Measured by comparing the wording of pages across versions.Interpretation. A reading of the evidence, not a direct observation.

Four generations of diagrams

The diagrams went through four technologies:

  1. SVG drawings, the sources of the Sphinx site’s figures.
  2. The same SVG files carried into the Astro site, with translated copies for the English pages from September 21, 2025.
  3. Mermaid diagrams: tried on June 10, 2026 and reverted the same day, then adopted from September 3, 2026.
  4. Native Astro components that draw each diagram as SVG when the site is built.

Not every diagram went through every stage: some pages still use SVG files.

How this is known: Git. Observed in Git: a commit, its date, a tag, a branch, a file or its contents.

How the history was reconstructed

Every statement on this page carries one or more of these markers. Select a marker to see what it means.

Evidence categories

GitObserved
Observed in Git: a commit, its date, a tag, a branch, a file or its contents.
Content comparisonObserved
Measured by comparing the wording of pages across versions.
Author’s wordsReported
The author's own words: a commit or tag message, the site's own text, or a fact the author recorded by hand in this history, such as the domain registration.
InterpretationInferred
A reading of the evidence, not a direct observation.
UncertainNot established
Not established by the evidence.

Method

Git activity
The first and last commit recorded for a platform on its own line: master for Sphinx and Astro, the branch for the DITA experiment. Earlier work on a side branch is shown separately. These are not production dates.
Production dates
Production dates are when a platform served the public site. Git activity does not prove deployment, so where the evidence is silent they are shown as not recorded.
Quiet periods
An interval of at least 240 days between two consecutive commits on master, the production line, ordered by commit time. It says nothing about whether the site was online.
Content matching
Each page of the last Sphinx state is matched by its wording, not by its file name, to the 2014 import, the DITA branch, the Markdown converted in September 2025, and today’s French page. Git does not record the 2025 conversion as renames, so file names alone cannot link the versions.

Technical and source details

How the figures are computed, and every historical page with where its text is today.

How scores and derived figures are computed
Scores
Each score measures how much of an earlier text is retained in a later one: the share of its five-word sequences (5-word shingles) found in the later text. Markup changes lower the score, so read it as approximate.
Derived figures
Some figures are computed from Git rather than read from it. For the age of pages at the 2025 conversion, a commit counts as an individual edit when it touches 20 pages or fewer. Counting only commits that touch a single page gives an older median. The pandoc match compares files with line wrapping ignored. Git records parts of the conversion toolchain, not every step.

Explore the lineage

Every historical page and where its text ended up. Historical titles are in French, the language the site was written in.

kept as its own page39at the 2025 conversion

Kept a page of its own through the 2025 conversion, and still has one.

folded into another page62at the 2025 conversion

Has no page of its own after the conversion. At least 30% of its text is found inside another page.

  • text now in About Olivier Carrère (4)

    • Formats cibles
    • Formats sources
    • Les sources de ce blog sont gérées sous Git
    • Libérez vos informations de leurs silos
  • Documentation technique : diminuer les coûts, améliorer la satisfaction clienttoday: Technical documentation: reduce costs, improve customer satisfaction
  • Format structuré DITA XMLtoday: Structured DITA XML format

  • text now in From document to modular document base (6)

    • Gérer son contenu DITA XML avec ou sans CMS ?
    • Le single-sourcing : un format source, plusieurs formats cibles
    • Les topics, modules d'information de base DITA XML
    • Organisation à la demande du contenu
    • Typologie de haut niveau de l'information
    • Un langage à balises

  • text now in Case study: NuFirewall documentation (8)

    • Centraliser les conref dans un fichier unique
    • Fournir une information ciblée avec le texte conditionnel ditaval
    • Imbriquer les conref
    • Maximiser l'utilisation des conref pour faire baisser les coûts
    • Partager des blocs d'information atomiques avec les conref
    • Prendre en compte les contraintes de traduction
    • Protéger les informations confidentielles
    • Utiliser le nœud XML de plus bas niveau

  • text now in Structured and unstructured formats (5)

    • DocBook ou DITA XML ?
    • Migration de FrameMaker vers DITA XML
    • Migrer de FrameMaker vers DITA XML
    • Restructuration du contenu FrameMaker
    • Table de conversion FrameMaker vers DITA XML

  • text now in Git: from file to content (2)

    • Faire sauter les goulets d'étranglement avec les branches
    • Organiser son historique avec Git rebase
  • Rédaction technique : un processus industrieltoday: Technical writing: an industrial process

  • text now in Source format (3)

    • Documents monolithiques ou modulaires
    • Fichiers binaires ou texte
    • Qu'est-ce qu'un module d'information ?
  • Workflow de création et validationtoday: Validation and quality control

  • text now in DITA XML and XSL-FO tutorials (10)

    • Accélérer sa saisie avec le mode Predictive pour Emacs
    • Afficher un index dans un PDF (mais pas sous DITA Open Toolkit)
    • Créer des documents différents à partir des mêmes sources DITA XML (texte conditionnel)
    • DITA Open Toolkit : afficher les références croisées dans les PDF
    • Générer un PDF avec DITA Open Toolkit sous GNU/Linux
    • Générer un PDF avec DITA Open Toolkit (Windows)
    • Gérer les projets de documentation multilingues DITA XML
    • Utiliser l'IDE nXML pour DITA XML
    • XSL-FO : filtrer du contenu selon des conditions « sauf » et « ou »
    • XSL-FO : insérer automatiquement un titre pour les exemples

  • text now in Managing a project from start to finish (15)

    • Communication interne
    • Définition des rôles des membres du worfklow photo
    • Définition d'un workflow de gestion des photos
    • Didacticiels vidéo
    • Galerie photo centralisée dans le cloud
    • Gestion des photos d'une association
    • Iconographie
    • Indexation des photos
    • LaTeX
    • Patience…
    • Récupération et tri des photos
    • Sauvegarde incrémentale et décentralisée
    • Style rédactionnel
    • Suivi des modifications sous Git
    • Support papier

  • text now in The Raspberry Pi 3 as a documentation platform (5)

    • Configurez le Raspberry Pi 3
    • Créez et modifiez les schémas
    • Générez votre documentation
    • Gérez les versions de votre documentation
    • Installez les logiciels nécessaires à la gestion de ce blog
not carried over6at the 2025 conversion

Less than 30% of its text is found anywhere in the converted or current site.

No page today

  • Contact
  • Créez et modifiez le texte
  • Une documentation en microservices
  • Persona : à quoi ressemblent vos utilisateurs ?
  • Le coin du geek
  • Récupérez les sources de ce blog
kept, later rewritten1after the conversion

Kept a page of its own in 2025, but today’s page retains less than half of the text the conversion carried over.

carried over, later deleted2after the conversion

Converted in 2025, then deleted.

No page today

  • Jourdainisation en ligne d'une tirade ·
  • Un index est-il utile dans un PDF ? ·
dropped in 2014 clean-up38

Imported from WordPress and deleted in the October–November 2014 clean-up, before release 1.0. It never reached the later Sphinx site.

No page today

  • Animation Flash : partez à l'assaut du grand public ·
  • Appauvrissement volontaire du discours ·
  • Concept DITA XML ·
  • De la modularisation au partage de l'information ·
  • Des schémas, mais pas de captures d'écran ·
  • DITA Open Toolkit ·
  • DITA XML ·
  • Exemple d'animation Flash ·
  • Exemple : documentations d'une version libre et propriétaire d'un même logiciel ·
  • Faciliter le travail de groupe avec Subversion, cron et Emacs ·
  • Formats structurés ·
  • Guide de l'utilisateur : aidez vos clients, séduisez vos prospects ·
  • Les balises sont sémantiques ·
  • Livre blanc : mettez le contenu technique au service de votre communication marketing ·
  • Logiciel de gestion de versions ·
  • Logiciels de gestion de versions ·
  • Magazine d'entreprise : fidélisez vos clients et augmentez votre couverture médias ·
  • Minimalisme - moins, c'est plus ·
  • Modularisation du contenu ·
  • Outil de gestion de contenu d'entreprise ·
  • Partager des unités d'informations atomiques avec les conref ·
  • Placer le maximum d'information dans la structure ·
  • Plaquette commerciale, brochure ou flyer : rendez votre technologie attrayante ·
  • Rédaction structurée ·
  • Reference DITA XML ·
  • Répertoires ·
  • Séparation du format source et du format cible ·
  • Site web : convainquez tous vos publics, du décideur à l'expert ·
  • Spécialisation DITA XML ·
  • Structuration et hiérarchisation de l'information ·
  • Structure de table des matières : map ·
  • Task DITA XML ·
  • Tests de non-régression ·
  • Texte conditionnel ·
  • Typologie de l'information DITA XML ·
  • Un format adapté aux entreprises de toutes tailles ·
  • Workflow ·
  • XSL-FO : insérer du texte à partir des fichiers DITA ou des fichiers de variables ·