clement13.net
Original photo: A prism splitting light
Description
A prism splitting light
Source
Kelvinsong · original page
Licence
CC0 Public domain
Original photo, unretouched. The background is a version of it stylised by the site.

How This Site Is Built

A site is also judged by how it's maintained. Here is the pipeline that produces it, the checks that keep it from degrading, and what it deliberately chooses not to do.

≈ 3 min readReading
In brief
  • One source, one pipeline: pages and settings in plain text, a single generator, two languages, one archive per version.
  • Blocking checks: a broken link, a diagram with two overlapping boxes, or a tooltip that's too long prevents the build.
  • Published metrics: the Site Health page shows the current status and how it evolves, including what still remains to be done.
On this page

The pipeline, from source to archive#

Content is written in plain HTML, with settings in JSON. A single generator assembles pages, draws diagrams, produces both languages, and checks the result. Nothing is rendered in the reader's browser, except a few interactions (menu, tooltips, dashboard charts).

SourcesHTML pages,JSON settingsGeneratorpages, diagrams,images, two languagesCheckslinks, layout,tooltips, translationsPublished sitestatic filesMetricslink crawler,size, qualityArchivesfull packageand published content
Each step can stop the pipeline: a detected issue never goes live.Hover over or tap an element to read its details.

What is checked, and what it triggers#

What is checkedWhenEffect
Internal links and anchorsat every buildthe build fails
Diagram layout: overlapping boxes, text wider than its frame, arrow label impossible to placeat every buildthe build fails
Hover tooltips: length, existing targetat every buildthe build fails
Translations: French changed since the translationat every buildthe page is flagged as outdated
Inline style attribute, multiple main titlesat every buildwarning
External links: page not found, missing domain name, invalid certificate, server error pageon demand, by the crawlerlist published on the health page
Page size, single main title, description, image alt texton demand, by the crawlermetrics published, by version
A link blocked to robots is not a dead link

The crawler distinguishes a dead link (not found, missing domain, invalid certificate) from an unverifiable link (programmatic access forbidden, anti-bot protection). The latter is never published as broken: being wrong by claiming a page no longer exists is an editorial error.

What the pipeline deliberately chooses not to do#

  • No external dependency for the reader: no content delivery network, no remote font, no tracker. Chart libraries are hosted with the site.
  • No inline script or style: content security policy forbids them, and the build checks for the absence of inline style.
  • No cookie: only anonymous audience measurement, hosted on the site's domain, counts page views.
  • No hidden date: neither in pages nor in archives. Only content carries dates, and versions are used only as numbers.
  • No character-drawing diagrams: diagrams are text-described objects (boxes, arrows, labels, tooltips), redrawn in the site's visual style, light and dark.

Measure to improve#

The Site Health page does more than provide a snapshot: a version filter reveals changes in volumes, translations, checks, illustrations, and diagrams, with a “coverage” area that grows (or not) from one version to the next. The version log gives the details, topic by topic.

What these metrics show is not always flattering: the share of checked pages drops when new guides arrive before being reviewed. It is published as is.