This Month in React

The podcast's content pipeline and website. Episode outlines and transcripts live as markdown in content/episodes/; everything published — the site, the RSS feed, the chapters, the newsletter body, the search index — is derived from those files at build time.

Setup

cp .env.example .env   # fill in what you need; every variable is optional for the site
npm install
npm run dev            # http://localhost:3000

Site

TanStack Start, fully prerendered to static files. There is no server runtime in production: dist/client is the whole deployment.

Two rendering paths share one Document component and one stylesheet:

• Static documents — /episodes/<yyyy-mm> and /links. A server route handler renders the whole page with renderToStaticMarkup. Content appears once in the HTML, there is no hydration, and the only JavaScript is a short inline script (timestamp seeking; link filtering).
• Router pages — /, /about, /search. Ordinary prerendered routes with the framework runtime, about 109 KB gzipped. RSC is enabled in the build for future supporting pages.

Transcripts are deliberately not rendered through RSC or a router loader: both inline a second serialized copy of the content for hydration, which doubles an hour-long episode page.

Commands

Command	Effect
npm run dev	dev server on http://localhost:3000
npm run build	prerenders every route to dist/client, then runs Pagefind
npx pagefind --site dist/client	rebuilds the search index by hand (postbuild does this normally)
npm run check	asserts the build output and prints sizes
npm test	unit tests

Environment variables

Every variable in this repo — build variables and script credentials alike — lives in one gitignored .env at the repo root. Set it up with:

cp .env.example .env

.env.example is committed and is the authoritative list, one comment per variable; this README does not duplicate it. Vite reads .env natively for the VITE_-prefixed variables, and the npm scripts pass Node's --env-file-if-exists=.env for the rest. There is no dotenv dependency.

The site's variables are all optional and all read at build time: VITE_SITE_NAME, VITE_SITE_URL, VITE_BUTTONDOWN_USER, VITE_ATPROTO_PUBLICATION_URI, VITE_ATPROTO_DID, VITE_BLUESKY_PROFILE_URL. Each one that is unset simply omits its feature — the newsletter form, the /.well-known file, the comments section, the Bluesky link.

Set the same values in Netlify's build environment (.env is not deployed). VITE_SITE_URL must be set before launch or the feed will carry placeholder URLs; it is also what publish-atproto writes as the publication url, so the two can never disagree.

Publishing an episode

1. Record and edit in Descript; publish the audio to Transistor.
2. Write the outline in content/episodes/<yyyy>-<mm>.md — a nested list of topics, each with a link and a timestamp. That one list becomes the table of contents, the chapters, the link index and the newsletter body.
3. npm run publish-transcript <yyyy-mm> <descript-project-id> — appends the speaker-labelled transcript and pushes an SRT to Transistor.
4. npm run ingest — fills in audio URL, duration, season, episode number and people from the Transistor feed.
5. Commit and push. Netlify builds the site, prerenders every page and regenerates the search index, link index and RSS feed. Buttondown picks up the new /feed.xml item and mails the outline to subscribers.
6. npm run publish-atproto -- <yyyy-mm> — writes the episode's site.standard.document record to the show's AT Protocol account, links it to the announcement post, and stores the record's AT URI in front matter as atUri. Commit that one-line change.

AT Protocol

The site is a standard.site publication. /.well-known/site.standard.publication carries the publication's AT URI, each episode page carries <link rel="site.standard.document">, and — when the episode has an announcement post — a "Reply on Bluesky" link plus a comments section that fetches the reply thread from the public Bluesky API with a small inline script. That fetch is the only external request an episode page makes, and the page reads fine without it.

Routes

/ · /episodes/<yyyy-mm> · /links · /search · /about · /feed.xml · /episodes/<yyyy-mm>/chapters.json

Every one is a static file after the build. A broken internal link fails the build, because the prerenderer crawls <a href> from /.

Measured sizes

From a full production build of all 39 episodes on 2026-09-08 (npm run build && npm run check):

Output	Raw	Gzip
Router-page JS, all 5 chunks	348,891 B	111,943 B
Largest episode document (2025-12)	202,568 B	52,362 B
Newest episode document (2026-07)	140,274 B	37,407 B
/links	517,905 B	77,456 B
/feed.xml	192,573 B	55,130 B
Pagefind index (dist/client/pagefind, on disk)	2,609,725 B	—

Only the three router pages load the JS; episode and links documents reference zero /assets/*.js — npm run check asserts that on every episode, along with each transcript appearing exactly once.