The HTML is rendered on the server. Titles and descriptions are unique per route (lib/seo.ts). The canonical URL is SITE_URL plus the path. Open Graph and Twitter cards use the same copy and a generated image (app/opengraph-image.tsx, re-exported as the Twitter image). lang is en on <html>. Unknown paths return 404. /gallery and /gallery.html redirect 308 to /.
JSON-LD is only what the page shows: Person and WebSite with stable @id values ({origin}/#person, {origin}/#website), ProfilePage on the home page, BreadcrumbList where the breadcrumb is visible, Article on notes (the colophon’s byline is the site, not Ryan), and CreativeWork on project pages. No review, rating, or offer markup is emitted, because those facts are not on the page.
app/llms.txt/route.ts is a plain-text description of the site. It lists what is not claimed. It is not a promise that any model will read it.
Ryan is a public creator and wants to be found. app/robots.ts allows search and citation crawlers and disallows training crawlers.
Allowed, including /, and blocked from /admin, /portal, and /api/:
* (everyone else)Disallowed from /:
Google-Extended is a training opt-out token, not a separate crawler. Googlebot still crawls, and Google’s AI Overviews can still quote pages Google has indexed. Blocking Google-Extended does not remove the site from Google Search.
Perplexity-User is documented by Perplexity as generally ignoring robots.txt. This file does not pretend otherwise. It allows PerplexityBot, which is their search crawler.
Lighthouse 12.8.2 against next start, 23 October 2026, after the fix below: SEO 100 on /, /links, and /collaborate. Performance was 95, 97, and 95. Accessibility and best practices were 100. The full table is in docs/HANDOFF.md.
An earlier run the same day scored SEO 91 on those three pages. The only failing audit was meta-description. The description was in the HTML response and not inside the initial <head>. The shell called headers() to mark the current nav item, which forced every page to render dynamically, and Next.js streamed the metadata into the body. Lighthouse did not count it.
That call is gone. The current path comes from usePathname() in components/site-nav.tsx. Pages that do not read searchParams are static, and the description is in <head>. Routes that read searchParams (/open-source, /bookmarks, /mentorship/book, and the admin screens) still render dynamically, so their metadata can still stream. Admin and the portal are noindex. Those dynamic routes were not part of the Lighthouse sample.
See the last section of docs/deployment.md. Submit the sitemap only for the origin in SITE_URL.