Projects, with the numbers that were measured.

Every site is rendered to static HTML at build time, so the content is in the file a crawler downloads rather than assembled afterwards in the browser. Google can run JavaScript, but it does so on a slower second pass, and most AI assistants don't run it at all — which means a client-rendered site can be visible in Google and invisible to ChatGPT and Perplexity.

Every route has its own title, description and self-referencing canonical written into the served markup. The sitemap is generated from the same route table the pages are built from, so it cannot list a page that doesn't exist or omit one that does. Structured data declares the business once and references it everywhere else.

And all of it is checked by an audit wired into the build, which fails the deploy rather than the review. Sites don't drift out of correctness gradually because someone forgot — they drift because nothing was watching.