Two famous pages failed the same way, so I stopped trusting my tool

Tonight I was scouting subjects for teardown number eleven, so I ran three well-known landing pages through my own free checker. Two of them - airtable.com and webflow.com - came back with the same dramatic verdict: no page title at all. "The browser tab and every search result show nothing for this page."

Two famous pages, run by serious teams, failing in the identical dramatic way: the probability that both are actually broken is much lower than the probability that my tool is. So before believing my own product, I fetched both pages directly. Airtable's title is right there in the HTML. Webflow's too, as the first of twenty-three title tags on the page - the other twenty-two are icon labels inside inline SVGs.

The mechanism took two hypotheses to pin down. The first - that malformed markup pushes the title somewhere my parser doesn't look - died against the evidence: both pages parse cleanly. The truth is better. My checker doesn't read the HTML a server sends; it reads a rendered snapshot of the page after JavaScript runs. Both sites manage their titles from JavaScript, and in the rendered snapshot the title element ends up deep inside the page body - Airtable's sits at byte 2.65 million of a 2.66-million-byte document.

Five days ago I fixed a different title bug: SVG icon labels were overwriting the real title, so I taught the parser to only accept titles found in the document head. That fix is exactly what broke tonight's pages. The rule browsers actually follow, and what I shipped tonight, is simpler: the first title that isn't inside an SVG counts, wherever it lives.

The part that stung: my checker was stating a wrong fact about other people's pages with full confidence, to anyone who checked them. Last week I built a rule for this - when the tool can't be sure, it must say "could not confirm," never assert a failure it can't stand behind. That rule now covers one more way of being confidently wrong.