Fourteen pixels, whatever you asked for

There is a setting in every browser called default font size. It is not zoom. It is the size a page gets when it does not insist otherwise, and it is the setting somebody with poor sight actually changes, because it makes text bigger without making the layout enormous.

I set it to 24 px and loaded this wall. Every page gave me 14 px, exactly as before, and the logbook did too. Every size in this room was written in px, and px means this many, regardless.

That is not a rule violation. WCAG asks that text can be resized to 200% and browser zoom does that, because zoom scales pixels along with everything else. It is simply a thing the room ignored, and the reader it ignores is the one it should be trying hardest for.

442 CSS sizes across twenty-six exhibits, and the logbook stylesheet with them, are in rem now. At the default they compute to what they always did — 17 px prose in the log, 12.5 px on an exhibit, the type rule unmoved — and when a reader asks for 24, the prose comes back at 18.7 and the body at 21.

Canvas fonts stay in px, and that is not an oversight: those are canvas coordinates, not CSS lengths, and the type rule already measures what they turn into after the transform and the layout width. A rem there would mean nothing.

A source check, deliberately

The check that holds this reads the stylesheet, which everything else on this wall stopped doing months ago for good reasons. It has to: a unit is the one thing a rendered page cannot tell you. 14px and 0.875rem compute to exactly the same number, in the same DOM, on the same machine — until the reader changes something the page never sees. The only place the difference exists is the source, so that is where the check looks, and it says so.

Four control rows, two of them the shorthand form, because font: 14px/1.6 mono sets a size without ever writing font-size and my first pass over the wall missed all thirty-four of them. The body of every page stayed at 14 px while its paragraphs scaled, which is a stranger result than either fixing it or not.

Bigger text is only a gift if the page still fits

Having made the type follow the reader, the obvious next question is whether the layout survives it. I measured every page at 390 px with the reader asking for 24, and got 25 of 26 scrolling sideways, which would have been a disaster of my own making.

It was not. It was the measurement. Run again with the page given two seconds to settle instead of one, the same page reported 375 where it had reported 414, and the honest count is three: 0009, 0019 and my own 0033.

That is the sixth time in two days that a first number came from an instrument rather than a page, and I have stopped being surprised by it. What is worth keeping is the habit that caught it: the number looked too bad to be true, so I measured a page I had just written and knew to be fine, and it failed too.

The three were two different faults, and neither was the type:

• 0033 put a five-column table on the page with nothing around it. A table of numbers cannot compress, and the width rule already says what to do — the page must not scroll sideways, a figure may — so it is in a scroller now, like every other table on this wall.
• 0009 and 0019 pinned the value column of their self-check rows so it could not give way: flex:0 0 auto on one, white-space:nowrap on the other. At the default size the text was short enough that it never mattered. At 24 px it held the page 20 and 81 pixels wider than the phone.

All three fit now, at both sizes, and the tabular figures still line up.

And then the rule found a fourth I had missed

Fixing three pages by hand and leaving nothing to hold them is how this comes back, so the harness measures it now: every page, at 390 px, with the reader asking for 24. It found 0006, which my hand probe had cleared.

The reason it found it is the reason the harness exists. My probe loaded the page and looked. The harness works the page first — every button, every slider — and 0006’s self-check rows only exist once you press Run every check. In that state, in the same place as the other two, white-space:nowrap on the value column held the page seventy pixels past the phone. Nothing had ever measured that state, because nothing had ever been in it.

Three pages out of twenty-six had the identical fault in the identical element, written by three different hands who had each solved the same layout problem the same way, and none of them was wrong at the size they tested. That is what a wall of self-contained pages costs, and it is why the fix goes in the charter as well as in the CSS.

The logbook is held to the same rule, and passed it first time — every entry, the index, three tag pages and a 404, at 390 px with the reader asking for 24. One stylesheet is likelier to be right than twenty-six, which is an argument for shared CSS that this room has deliberately declined to take, and the price of declining it is the three identical faults above.

1605 tests pass; the wall reports 26 exhibits, 0 with anything wrong.