The site has a checker, sideways.mjs, that opens every published page in headless Chromium and asks one question: is the page wider than the window? A page that is gets zoomed out or scrolls sideways, and a visitor on a tablet sees the right edge of a calculator cut off. It had been run at phone widths and at 820 and 900 px. Today it ran at four widths it had never covered: 600, 700, 1,040 and 1,100 px, across all 8,939 pages.
The first answer was 6,230, and it was wrong
Almost every failure named the same thing as the widest box: the site header on the tool pages, 440 px too wide at 600 px. 1,040 minus 600 is 440.
The checker loads each page once at the first width, then resizes the window for the others. It waits one animation frame and measures. The header, though, stretches itself to the window from a small script that listens for resize and waits 60 ms before acting, so it isn't recomputed on every pixel of a drag. One frame after a resize it still had the old width. The default widths only ever make the window wider, so the gap had never shown up.
A visitor would see this for 60 ms when rotating a tablet. It is not a defect. The checker now waits 150 ms and measures again whenever a resized width comes out too wide, so clean pages cost nothing extra. Rerun: two failures.
The real ones
There were three real pages (the first run had found one of them already):
- The clothing design studio, from 901 to about 1,300 px: 161 px too wide at 960. Two inputs in the screen-print form had no
typeattribute, so the stylesheet'sinput[type=text] { width: 100% }never reached them. They kept the browser's default width and pushed the grid open. The dropdown next to them was squashed down to the letter "F". - The 4×8 vs 5×5 plywood comparison in the cut-list optimizer: 112 px too wide at 700.
- The job margin variance calculator: 101 px too wide at 700.
All three come down to one CSS rule that keeps catching people out. A grid track sized 1fr has a minimum of auto, which means "no narrower than my content". An input with no width set has a content width of roughly 150 to 270 px, depending on font and padding. Put three of those in 1fr 1fr 1fr and the grid can't get narrower than about 800 px, whatever the window does. The fix is to write the track as minmax(0, 1fr) and let the input fill its cell (min-width: 0; width: 100%). Then the column decides the width, not the input.
Each page is now 0 px too wide at 14 widths from 320 to 1,280, and each has a test that pins the fix. The checker change matters more for the long run: a sweep that cries wolf 6,000 times gets ignored, and these three pages would have stayed hidden underneath.
Details and screenshots for the next run: docs/reports/polish-2026-11-12-wide-tablet-sweep.md.