What I did to make this site findable, and what the numbers could not show

The problem

A new site has no history for a search engine to trust, so the page itself has to say what it is. My first build looked right and scored well, but when I counted, none of its seven pages said which address was the real one, none had a sharing preview, and there was no sitemap or robots file at all.

What I changed

• A canonical tag and Open Graph tags on every page. Each page now names its own address and carries a title, description and preview for when someone shares it.
• A sitemap and a robots file. The sitemap lists only pages that exist, and the build fails if one points at a missing file.
• More structured data. JSON-LD went from 2 blocks to 4: the Person block gained an @id and a URL, Home gained a WebSite block and About gained a ProfilePage block.
• The four practice sites kept out of search. The demo sites that sit under this one carry a noindex tag on every page (159 pages), so they cannot compete with the portfolio. The tag, not a blocked path, because a crawler has to read a page to see it.
• Checks that fail the build. A missing canonical, a JSON-LD block that does not parse, or a sitemap entry with no file stops the publish step.

What the measurements showed

I ran Lighthouse 12 on the live pages on 5 October, and compared the first build with the current one on my own computer, with the same settings and the median of three runs. Every live page scored 99 or 100 in all four categories. On a simulated slow phone connection, the largest element on Home appeared in 1.8 seconds, and the page transferred 115 KB.

Two things went wrong on the way, and I kept both in the record. The accessibility score of one page was 96 because links were told apart from text by colour alone; underlining them brought it to 100. And my first draft of this page said every page scores 100. The live re-run showed one page at 99 on mobile, so I corrected the claim before publishing.

Speed did not improve. The comparison shows mobile performance at 99 for the first build and 98 now, because the site gained code and tags. So the accurate word is held, not improved.

What the tool could not show

Lighthouse gave the SEO category 100 both before and after, so the score cannot show any of this work. That is why I counted the tags instead. On 7 October, Google's site: search listed 2 pages of the site (Home and About) and no case study yet, while Bing listed about 50 results. I resubmitted the sitemap to Search Console. There is no field data from real visitors, because a site this new has none, so I make no claim about rankings or traffic.

What I would do differently

I would add the canonical tags, sitemap and structured data in the first build, since they cost almost nothing when the pages are being written and they took a second pass to add. I would run Lighthouse on a real phone and on Chrome, not only on an emulated one in Edge. And I would measure the pages that I only ran once more than once, because I do not know how much those scores move from run to run.