404 / missing route
Read · detail. A not-found page should name the failure plainly and give the reader one known way forward.
Open original (Vercel, verified 30 Aug 2026; HTTP 404)
Original (cited)
Live URL: vercel.com/docs/errors/does-not-exist
The official route responds with HTTP 404. Its visible not-found content is intentionally small; this block map records the observed reading order rather than reproducing the source page's skin or assets.
404
This page doesn’t exist.
Return Home
Docs · Knowledge Base · Blog
The cited page offers one recovery link (Return Home) plus links to the Vercel Docs, Knowledge Base, and Blog. The missing-route message and lightweight recovery navigation are the structure we are borrowing.
Gelium remake
This is how that structure lands in Gelium: the same status-first message, with our recovery copy and a real GET link back to the section-reference index. The component is the registered server-rendered ui-error-state; it is not a screenshot, a Vercel clone, or a link to an unrelated recipe.
Page not found
This page doesn’t exist. Return to the section references index to continue.
Back to section referencesKeep / adapt
| Kept | Gelium primitive | Rejected | Why |
|---|---|---|---|
| Visible 404 status | ui-error-state-code |
Vercel logo or product chrome | The reference is a not-found state, not a brand shell |
| Plain missing-page message | ui-error-state-title + ui-error-state-body |
Copied wording as product copy | The job is shared; the voice belongs to Gelium |
| Single missing-route composition | ui-error-state with role="alert" |
Decorative illustration or animation | Adds skin without improving recovery |
| Recovery destination | real .ui-button GET link |
Dead or JS-only retry control | The user needs a known, keyboard-accessible next step |
Ask before copying
- Should a missing route return to the documentation index, the product home, or the user's previous known context?
- Does the product need a support or status link for failures beyond a simple 404?
- Should the page retain the product shell, or is a focused error state safer for the failure context?