The site for Felix. The blog is built with Zola; every other page is a plain HTML file.
| Path | What it is |
|---|---|
static/ |
Copied to the site root as is: the pages (index.html, features.html, ...), assets/, icons, CNAME. .github/scripts/release_versions.py keeps the release markers in static/index.html and static/projects.html current. |
content/blog/ |
One Markdown file per post, plus _index.md for the blog index. |
cards/ |
The illustration shown on each post's card in the list. |
templates/ |
The blog's page shell, header, footer, cards, feed and sitemap. |
static/assets/img/felix-yarn.svg |
The full logo: the cat batting a ball of yarn, with the wordmark. The home page hero carries an animated inline copy of it (styles in static/assets/home.css; it plays once, then the arm and yarn ball swing every 7 to 11 s while the logo is on screen); the PNG and WebP files beside it are exports for places that can't use SVG. The nav and favicon keep the head mark. |
.github/scripts/notfound_art.py |
Writes the animated figure into static/404.html. Change the drawing or its timing there and rerun it; don't edit the generated markup by hand. |
Install Zola (brew install zola; CI uses the version pinned in
.github/workflows/pages.yml), then:
zola serveand open http://127.0.0.1:1111/blog/. To build the output that gets deployed,
including the blog/<slug>.html flattening, run .github/scripts/build.sh and then
python3 .github/scripts/check_links.py. The output is in public/.
Create content/blog/<slug>.md. The slug is the URL: the post is served at
/blog/<slug>.html.
+++
title = "How we catch a 1% regression"
description = "One sentence. Shown under the title, on the card, in search and in the feed."
date = 2026-10-17
slug = "catching-a-one-percent-regression"
path = "blog/catching-a-one-percent-regression.html"
[taxonomies]
topics = ["performance", "testing"]
[extra]
read_minutes = 7
card = "cards/catching-a-one-percent-regression.svg"
+++datesets the order, the byline and the feed'spubDate. Posts on the same day are ordered by a time in the date (2026-10-17T12:00:00, later first); the time is not shown.pathmust beblog/<slug>.html. Keep the.html: it is the address readers and the feed already use.read_minutesis the word count divided by 230, rounded.topicstakes one to three names. Use the existing ones where they fit (architecture, compatibility, durability, networking, performance, replication, testing); a new name creates a topic page and a topic feed.cardis an inline<svg>file for the list page: the post's first figure withrole,<title>andaria-labelledbydropped,aria-hidden="true"added, and everyid(and itsurl(#...)/href="#..."references) prefixed so ids stay unique on the list.
Write the body in Markdown: paragraphs, ## and ### headings, lists, links and
`code`. Headings get an id from their text; to keep a shorter one, write
## Title {#short-id}. Anything Markdown cannot express stays as raw HTML in the
file:
- Figures:
<figure class="fig ck">with an inline SVG, andclass="fig ck wide"for a figure wider than the text. The shared illustration colours and text styles are instatic/assets/ck.css. Do not leave a blank line inside a raw HTML block; Markdown ends the block there. - Highlighted code:
<pre class="code" tabindex="0"><code>withtok-*spans (class="code tla"for TLA+). Plain fenced code blocks also work. - Callouts and tables:
<aside class="callout">,<div class="table-wrap">.
content/style-sample.md (served at /blog/style-sample.html, noindex, not in
the list, feed or sitemap) shows each of these.
The <head>, header, footer, topic chips, the feed and the sitemap come from the
templates, and asset URLs get their ?v= cache key from the file's content.
Nothing else needs editing when a post is added.
The blog index searches title, summary and text in the browser, using Zola's
search_index.en.js and the vendored static/assets/elasticlunr.min.js
(elasticlunr 0.9.6, from
https://cdnjs.cloudflare.com/ajax/libs/elasticlunr/0.9.6/elasticlunr.min.js).
Both load when the search box is first focused.
.github/workflows/pages.yml builds the site on every push and pull request,
checks the links in the built output, and on main deploys it with
actions/deploy-pages. The repository's Pages source must be set to
"GitHub Actions".