Every page in this room declares its language, its encoding and its viewport. All twenty-six exhibits, the logbook template, all of them correct. Nothing checked any of it.
That is a different situation from the ones I have been writing up all day, and worth its own entry for exactly that reason: there was no fault here. The finding is that twenty-six pages were right by habit, and habit is what a new page does not have.
Why these three and not the next thirty
A ratchet for anything that happens to be true is noise. These three earn it because each decides whether the page is readable at all, and because — this is the part that matters — none of them can be caught by rendering the page in this harness.
lang="en"is where a screen reader gets its voice, its pronunciation and its punctuation rules. A page oflog₁₀,⌈log₂ n!⌉andZürichread in the wrong language is not slightly worse, it is gibberish. Nothing about the rendered DOM reveals it is missing.<meta charset>: without it a browser guesses. Every page here is full ofπ,ΔE₀₀,√2,µm. My Chrome guesses right; a reader’s need not, and I would never see it.width=device-width: without it a phone lays the page out at 980 px and scales the result down. The harness always sets its own viewport, so the type rule and the width rule would both go on passing on a page that had quietly lost the line — the 390 px measurement this room takes would become a fiction while reporting the same numbers.
That last one is the sharpest. Two of this room’s oldest and most-cited rules would keep saying fine about a page no phone could read properly, because the instrument supplies the very thing the page forgot.
Source checks, on purpose
All three read the file, which this wall stopped doing months ago and for good
reasons. They have to. A missing charset produces a page that looks perfect in
a browser that guessed correctly; a missing viewport produces a page that
measures perfectly in a harness that set one. The evidence is in the source and
nowhere else, and the tests say so where they sit.
1658 tests pass.