Skip to content

feat: phaethon.app as one react page on the app's own components, with one shader and less copy - #10

Merged
jerynv merged 4 commits into
mainfrom
feat/site
Nov 17, 2026
Merged

jerynv merged 4 commits into
mainfrom
feat/site

Conversation

@jerynv

@jerynv jerynv commented Nov 17, 2026 •

Copy link
Copy Markdown
Owner

phaethon.app is live, but the first version was a hand-rolled vanilla page with art everywhere: a shader in the hero and another in the closing, gradient tiles on every feature card, a dot matrix, an eyebrow pill, and too much copy. It also didn't use any of the app's actual components.

This rebuilds it as one React page on the app's own shadcn (Base UI) components, tokens and fonts, imported through the @ alias, so the site and the app share one design system. There's one shader left, behind the app screenshot. The rest is type, hairlines and cards. The copy is cut down to what someone needs to download it and know where the project stands.

The build still reads the latest release, CI state and last commit from GitHub and prerenders them, so the page is complete with no JavaScript. In the browser it hydrates, picks the visitor's OS for the download button, and refreshes the data live. The open issues count no longer includes pull requests.

Tested in headless Chromium against the production build: light and dark at 1440 and 390 px with no hydration or console errors, axe clean, and no horizontal scroll. A Windows user agent gets the .exe button and the Windows card. Copy puts the clone command on the clipboard. Switching the system theme flips the page. With JavaScript off, the real download links and status rows are still there. Losing the WebGL context falls back to the CSS art. pnpm typecheck, pnpm test and pnpm build pass.

It also fixes a flaky e2e that failed this PR on macOS: the long-conversation test assumed the closing diagram always draws within the app's 600 ms reveal wait, which a slow runner misses (the reply then grows by 202 px after it shows). It now accepts that case but checks the wait was kept in full, and still checks nothing moves once the diagram has drawn. Reproduced locally by cutting the wait to 1 ms: the old test fails with the same 202 px, the new one passes, and the normal path passes 3 of 3.

And one more flake that failed it on macOS: the 1500-conversation sidebar test timed frames from outside (p95 118 ms against a 66 ms budget, with zero long tasks, so the runner, not the app). It now counts long animation frames, the page's own script, style and layout time per frame, with the same budget. Locally it passes 3 of 3 at 0 slow frames, and an 80 ms busy loop on scroll fails it with 36 of 39.

…a commit message literal, and shows the css art if webgl drops
…he app's bounded reveal wait ran out on a slow runner, and checks the wait was kept
… animation frames, not time between frames a busy runner stretches
@jerynv
jerynv merged commit 2e47f85 into main Nov 17, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant