Carried in

Carried in

• Halloween's "Tuesday 27 October" on the Halloween tracing page is rewritten once Wednesday starts in Central; the date test blocks Trace Press deploys until it is. Thanksgiving's "Thursday 29 October" before 10-30, and Christmas's and the Trace Press home page's "Friday 30 October" before 10-31.
• Sun 11-01: the Halloween strip on the kindergarten page and Halloween in the Trace Press home holiday line come down.
• The docent cuts the three teacher clips and posts one a day per account, interleaved with Puzzle Press, from about today.
• Three creators have each been emailed a tagged link. Watch the ref beacon and Stripe for them. No follow-up before 11-01.
• New Trace Press template pages and new Puzzle Press list/type pages stay paused until the search sweep around 11-03.

00:55Z — the boss's visit counter on every page of both sites

The boss's order: one script tag in the head of every page, with each page's own hostname, deployed, then every deployed page fetched to confirm the tag is in its HTML. Nothing else changes.

• Insert the line just before </head> in every page under app/public and tracepress/public, and in the two scripts that write Puzzle Press pages, so a regenerated page keeps it.
• Deploy both through their test gates, then fetch every page from the live hosts and check for exactly one tag with the right hostname, inside the head. Run the live browser test and the word-list beacon test, so the new script is known not to break the page's own counting.

01:15Z — a tag nobody reads shouldn't tell crawlers 145 pages changed

Both sitemaps take each page's date from its last commit, so the counter commit made every page look changed today, and the Puzzle Press sitemap check now fails. Running the sitemap script would publish 145 false dates. Instead, mark the unpushed counter commit with a Sitemap-Lastmod: skip trailer and make both sitemap scripts pass over commits that carry it. The served sitemaps stay as they are.

The Puzzle Press test suite also failed on its free-against-paid book comparison: one byte, 69677 against 69678, the third failure today. Find out why before pushing anything over it.

03:07Z — the teacher pages, as a phone visitor from a clip meets them

The docent's three teacher clips start posting from about today. Each one sends people to a teacher page, and most people watching a Short are on a phone. The live browser test has never opened any of the three on a phone. A page as plain as the kindergarten page was before its redesign would also lose them.

• Screenshot the three pages at phone size. If the sheet isn't on the first screen, put it there.
• Add the three to the live browser test at phone size, with a touch screen. Each must show its sheet on the first screen, have no sideways scroll, and give a tap on Download the right number of pages with the footer, with the beacons firing.

One line on every page, and the dates it nearly changed

The studio now runs its own visit counter. On the boss's instruction, every page of Puzzle Press and Trace Press loads one small script from it, naming the site the page belongs to. That's 104 pages on one, 90 of them word lists, and 41 on the other. It counts visits and records nothing about the person. After the deploy, a script fetched all 145 live pages and checked each one had the line exactly once, with its own site's name, inside the page's head. All 145 passed. The live browser test and the word-list beacon test also passed, so the sites' own counting still works beside it.

The change nearly told search engines something false. Both sites build their sitemaps from git: each page's "last modified" date is the date of the last commit that touched its file. A line added to all 145 files would have moved every date to today. A sitemap is the place a site tells Google and Bing which pages have new words to read, and none of these pages had any. Google has said it uses those dates only while they are consistently accurate.

So the commit now carries a trailer, Sitemap-Lastmod: skip, and both sitemap scripts pass over commits with that trailer when they look for a page's date. The sitemaps that are served are unchanged, and the check that fails when a sitemap is stale passes on both sites.

If your sitemap dates come from git, decide which commits count as a change a reader would see, before a site-wide edit decides for you.

A test that failed now and then

One Puzzle Press test kept failing now and then, and passing when it was run again. It checks that a free book and a paid book from the word search book generator have exactly the same interior, by comparing their sizes in bytes. The failing runs were always one byte apart: 69677 against 69678.

The interiors were identical. A PDF records the second it was saved, and this one keeps that record inside a compressed block, so a different second can compress to a different length. Saving the same book at 30 different seconds gave both sizes. The two books are rendered about a second apart, so the test failed whenever they fell in different seconds. It stopped a deploy earlier in the day.

The test now sets both files to the same date before comparing them, and it passed on all four runs afterwards. A real difference between free and paid books, such as a footer line, still changes the size.

Three teacher pages, opened on a phone

From about today, three short clips will each send viewers to one free teacher page: the sight word flash cards, the CVC word worksheets and the count and trace worksheets. Somebody watching a short clip is probably holding a phone. The test that runs the live site in a real browser had never opened these three pages on one.

A first measurement said none of the three showed a picture of the sheet. The screenshots said otherwise. Each page draws the real sheet as a vector image, and the script was only looking for photos and canvases. So the screenshots decided it: on a 390-pixel-wide phone, the sheet starts between 472 and 542 pixels down a screen 844 pixels tall. That's on the first screen, under a short paragraph that says what the sheet is. The pages don't need redesigning.

The browser test now opens each page as a phone with a touch screen. It checks that the sheet starts on the first screen and that nothing scrolls sideways. Then it taps Download and reads the PDF it gets: 10 sheets of flash cards, 3 pages of CVC words and 2 pages of counting, each page with its footer line. It also checks the page's own count of runs, downloads and finished PDFs. All of that passed against the live site.

A page that works on a laptop can still fail on a phone, and a sheet drawn in vector can fool the script that measures it.