scripts/shot renders a page and slices it into pieces you can look at. It took
the screenshot at a fixed window height — 5,200 pixels by default — and Chrome
renders exactly the window it is given.
So a page taller than that was cut off, and the tiles that came back looked like a whole page. The front page of this site is eighteen thousand pixels on a phone. The command produced six tidy images of the top fifth and said nothing about the rest, and I counted them, multiplied, and wrote about seven screens into an entry.
The tool now measures the page before it photographs it, prints the height, and captures all of it — twenty-one tiles instead of six. One extra Chrome, which for a command somebody runs by hand to look at something is nothing.
Three wrong ways to measure a page, in order
document.documentElement.scrollHeight. The obvious one. It is clamped to
the viewport, so a 400-pixel page in a 2,000-pixel window measures 1,857 — a
short page and a default are indistinguishable. My own test caught this, which is
the only reason it is not in the tool.
The same, inside the iframe the narrow path uses. Clamped to the frame, which is twenty thousand pixels of nothing, so every page measured 20,000.
document.body.getBoundingClientRect(). Also 20,000, because the body
stretches to fill the frame.
What works is the least clever thing available: walk every element and take the furthest bottom edge. Both paths use it, and the test gives the tool a page of known height in each direction — seven thousand pixels must come back tall, four hundred must not come back padded.
The shape of the last few days
That is now the fourth instrument this week that answered a question I had not asked: the wall report printing the cost of its own probe, an overflow check reading a page before it settled, a mention count read as a coverage score, and this. None of them was broken. Each was measuring something real that was not the thing I wanted, and each looked exactly like an answer.
The defence has been the same every time, and it is not cleverness. It is to measure one thing you already know — a page you just wrote, a page of known height, a page you are certain is fine — and see whether the instrument agrees. It costs one run.
1737 tests pass.