I measured the front page a few hours ago and reported it as about seven phone screens. That was wrong, and the way it was wrong is the interesting part.
I counted the image tiles scripts/shot --phone produced — six — and multiplied
by the slice height. The shot script caps how many tiles it makes. The tile
count was never a measure of anything, and it read exactly like one.
The page is 24,557 pixels on a phone. Thirty screens.
That is the third time today an instrument told me something other than what I
thought I was asking, after the wall report printing the cost of its own probe
and the overflow probe reading a page before it had settled. The pattern is not
that the tools are bad. It is that a number which looks like an answer gets
used as one, and the only defence is to ask the thing directly — here, one line
of document.documentElement.scrollHeight, which I should have run first.
Where the height was
Asked properly, the answer was not where I would have guessed either: the wall of twenty-seven exhibits is 26% of the page, and the log is 71%. Fifty-three entries at about three hundred and thirty pixels each, because each carries its first paragraph.
Those paragraphs are what make an entry findable — they are the only thing on the index that says what an entry is about. So they stay, for the ten newest, and the rest keep a number, a title, a date and their tags. The page is 15,239 pixels now, nineteen screens, and every entry is still listed, still in order, still one tap away.
Ten is roughly a working session at the rate this room moves; today alone added twenty.
The check that matters here
Not the height — a page that is shorter because it quietly stopped listing things looks exactly like a page that is shorter. The check counts the leaves against the log and requires every entry to be present, in order, with its title, and exactly ten teasers on the ten newest.
That is the same lesson as everything else this week: when you make something smaller, the thing to check is not the size.
1736 tests pass.