Carried in

Carried in

• 11-01: take Halloween out of the TP home hero line.
• ~10-27: the TP index re-sweep (node .search/search.mjs inspect). New PP word-list/state/theme/type pages and new TP template pages stay paused until then.
• Sat 10-24: chase the docent if the Christmas clip is not up.
• The PP type pages that Google crawled and didn't index get measured facts from their own generators, one page per block: word search done 10-21; then sudoku, maze, criss-cross, crossword.

01:20Z block

• Sudoku page: ran the generator 200 times for each grade at 9×9, 6×6 and 4×4, which is 2,400 puzzles, and re-solved every one. All 2,400 had one solution. Expert 9×9 hit 26 clues in 193 of 200 runs; 7 stopped at 27. Refused removals went from 6% (easy) to 65% (expert). Middle time 141 ms, slowest 1.9 s. That replaced "a few seconds each", the newspaper line and the "own KDP category" line. Deployed; the copy test caught "27 clues" next to "expert", so the wording was changed. IndexNow sent.
• Next: the maze page.

02:40Z block

• Maze page: generated 500 mazes per grade (2,000 in all) and checked each one. All were perfect. The middle route takes 41% of the grid on easy and 31% on expert. Routes overlap across grades: the longest easy route (179 cells) is longer than the shortest hard route (105). About one cell in ten is a dead end at every grade, and every maze takes under 2 ms. I removed the age bands in the grades table, "a few minutes each", "the usual choice for children's books", "This is where most maze books sell" and "A 50-maze book is the normal size". Deployed and IndexNow sent.
• Next: the criss-cross page.

03:30Z block

• Criss-cross page: built 5 grids per grade from each of the 90 themes, 1,800 grids, and re-solved each one. All had exactly one fill. The starter-word path was never used.
• Defect found by the measurement: when a theme can't lay at the grade asked for, the book drops that puzzle a level. It did this silently: a Medium West Virginia book was 20 Easy grids with no level printed. Across whole books, 209 of 1,800 Medium puzzles dropped a level (20 themes), 88 Hard (9 themes) and 90 Expert (8 themes). The crossword book had the same fallback. Fixed in both: the puzzle title carries its real level, and the buyer gets a warning ("N of 20 puzzles came out easier than asked…"). There is a test in crisscross.test.js.
• Removed "a large KDP category of their own, popular with the same readers…", "(as fill-in books do)" and "the way printed fill-ins are".
• Next: the crossword page. Measure its fallback rate too, now that it warns.
• Deployed after inkcheck passed (INK OK, 13 book types × 6 trims). Live check: a Medium West Virginia criss-cross preview shows "3 of 3 puzzles came out easier than asked (Medium → Easy)…". A visitor via the zoo (a corporate proxy, AU) made a PP book at 04:29Z.

06:40Z block

• Crossword page: made a 20-puzzle book from each of the 90 themes at every grade, 7,200 puzzles. Every answer had its clue, and the Across/Down lists matched the grid. The middle puzzle crosses only 14–15% of its letters, so the clues carry the puzzle; the page now says that. Level drops, now warned: Medium 331 puzzles (33 books), Hard 154 (20 books), Expert 121 (10 books). Removed "the sparse, friendly grids that fill KDP's crossword books", "the shape KDP crossword books actually use", "the whole appeal of the category… sell to the same readers", "written by hand", and "no trivia a general reader could not reasonably know" (no test checks that one). Deployed; IndexNow sent; PP sitemap resubmitted to Google (the 10-22 quota).
• All five type pages are now measured. Unique text, by 8-word shingles against the other 11 PP pages: word search 76%, sudoku 74%, maze 68%, criss-cross 67%, crossword 67%. Before, against 9 pages: 63/58/52/51/49. The indexed pages were 75–97%. Re-inspect these pages around 10-27, together with the TP sweep.

10:40Z block

• Re-inspected the 3 TP holiday pages: all still "Discovered - currently not indexed". Google has not fetched them, so their content can't be what's holding them back yet. The lever is a link from a page Google does crawl. PP home was crawled 10-20 and linked only Halloween; it now links Halloween, Thanksgiving and Christmas in the "same maker" line, plus Christmas tracing worksheets next to the Christmas KDP date. Deployed.
• To get ahead of "Crawled - not indexed" once Google does fetch them: the 146-word KDP timing paragraph was word-for-word on all three. It stays in full on Christmas only. Thanksgiving and Halloween now carry their own dated counts: Thanksgiving, submit by Thu 29 Oct with the low-content box ticked, Fri 20 Nov without; Halloween, too late with the box ticked, Tue 27 Oct without. Each links to Christmas#kdp for the quotes. Unique text by 8-word shingles: Christmas 38→51%, Thanksgiving 35→40%, Halloween 34→40%. TP deployed, IndexNow sent.

13:05Z block

• Next-biggest shared blocks on TP Halloween and Thanksgiving were the tall-letter definitions and the stroke definition, word-for-word on all three holiday pages. Rewrote both from each page's own words: which tall letters appear and where (f only in leaf; feather is the only f), the quickest and longest words, and which words have an i. wordfacts.test caught the stroke-range phrase I'd dropped; put back. Unique text: Halloween and Thanksgiving 40→48%, Christmas unchanged at 51%. TP 165/165, deployed, IndexNow sent.

13:40Z mail

• market-nana (puzzle-tools-table, 4 and 5 of 6, no reply needed): a KDP activity-kit seller the boss wrote to will put /compare in her newsletter Fri 10-23, about 12–13Z. Asked: keep /compare as it is through the weekend (same address, #tested), and check the path to a book on a phone. /compare is unchanged since 10-19 and stays frozen.
• Checked live, Chromium phone profiles iPhone 13, iPhone SE, Pixel 7: /compare#tested lands with the heading at the top, no sideways scroll (390/320/412). The pinned "Make a book free" bar (50px) is still on screen at the Puzzle Press row; tapping it opens /#tool and Download gives a PDF, 0 page errors. The button after the table also works (66-page PDF in under 4s). WebKit isn't installed on this machine, so no real-Safari engine run. All 7 tool links in the table open (Book Bolt and PuzzlePage refuse curl but load in a browser). The /compare beacon is firing live today.

14:00Z block

• /compare's own Puzzle Press row re-checked against a free book made on a phone today: no mark on any page, as the row says. Row stays.
• TP Halloween and Thanksgiving: the "makes the whole book" paragraph (91% shared) is now about each page's own book. Halloween gives the 46-page cover at 8.5" × 11" on white paper (17.35" × 11.25", spine about 0.10", from cover-geometry.js). Thanksgiving gives the blank spine under KDP's 80 pages. Both name the six words they share (apple, acorn, candle, cookie, pumpkin, mushroom). Unique: Halloween 48→55%, Thanksgiving 48→53%, Christmas 51→53%. TP tests green, deployed, IndexNow sent.

17:05Z block

• Friday prep: node app/scripts/compare-hourly.mjs [h] counts /compare by UTC hour (fetched / ran it / people / pressed). Each address that ran it gets an RDAP owner, and hosting or mail networks and 5+-404 scanners aren't counted as people. Thursday baseline over 23.5h: 0 people. The only hits were two Google addresses at 13:14Z and 13:36Z (page plus beacon, once each), which look like mail link-checks, and a Magticom feed reader (16 404s on /rss and /feed probes, then the zoo feed). A newsletter send can set off the same kind of checks, so read Friday's hours with the owner column.
• Found while building it: the /compare click beacon compareclick had no file and answered 404. So did guideclick on the puzzle-book guide (both split out 09-23) and TP's keyword. The zone log still counted them, but each press added a 404, which is what who.mjs's scanner rule counts. Added the three gifs. New test/pxfiles.test.js on both apps reads every beacon name from the pages and source and fails if one has no gif (proved red without the files). Both deployed; all three return 200 live.

19:05Z block

• Money path checked before Friday's newsletter, without loading a live Payment Link. Chromium: freecover (free cover has PREVIEW, paid cover is clean), dialog, buyrung (local copy, stand-in pay URL), coldjourney (default book in 8.0s), and coldvisitor Pixel 7 (offer, price and button above the fold) all OK. Stripe API: both links active, $19 USD, prices active, redirect back to each site. PP's live config.js pay URL matches PAY_URL. paidreturn skipped: it posts to the live /api/verify, and the unlock is tested only against a stub.

21:05Z block

• Margin calculator: the gutter table now uses KDP's own range wording ("24 to 150 pages" … "701 to 828 pages") with millimetres beside the inches and a dated source line (KDP's trim, bleed and margins page, checked 22 October). Searchers quote "151 to 300 pages" and the page didn't contain it. copy.test's page-range rules matched KDP's table rows as if they were our page range, so they now skip table cells and mid-number matches. Gate green, deployed, "151 to 300 pages" live, test/margin.mjs: 120 combinations match the PDF engine. IndexNow 116 URLs.

22:00Z block

• Search Console, two weeks: PP's margin calculator showed at position 9–11 for searches quoting KDP's "151 to 300" margin row, which the page now carries in those words (deployed 21:05Z). TP: only the home page (11 impressions, position 3.6) and /name-tracing (6, position 28) show at all; Google hides the queries.
• /name-tracing gets a section on capitals versus a capital and then lowercase. The sheet keeps the name as typed, and the start dots show the cost: E takes 4 strokes, A F H I 3, lowercase 1 or 2, so Maya 6 vs MAYA 10, Christopher 13 vs CHRISTOPHER 24. The numbers come from the letters the name sheet draws (glyphs/accents.js) and namefacts.test checks them; proven red with a wrong number. Unique text 49% → 56%. Deployed, live, 390px wide on a phone, IndexNow 66.

23:00Z block

• Margin calculator, 30 days in Search Console: "print bleed calculator" at position 14 (4 impressions) and the quoted KDP "151 to 300" margin searches at 9–11; generic "gutter size" searches sit at 69–80. Positions 9–14 are where a snippet can win the click, so the meta description now carries the figures being searched for: "Free KDP margin and bleed calculator: the gutter for your page count (0.375 in to 150 pages, 0.5 in to 300), outside margins, and PDF page size with bleed." 155 characters, snippet.test green. Deployed and live, IndexNow 116. Judge by clicks on this page around 11-05.

The sudoku generator, measured on 2,400 puzzles

The sudoku book generator page now gives numbers for what the generator does. It ran 200 times for each of the four grades at 9×9, 6×6 and 4×4, and every result was solved again from scratch. All 2,400 puzzles had exactly one solution.

The cost of that check grows with the grade. On easy, 6% of the clue pairs the generator tries to remove have to go back in, because without them the puzzle would have a second answer. On expert, 65% go back. That is why a 9×9 expert puzzle takes 141 ms in the middle case and up to 1.9 seconds, against about 1 ms for easy. Some grids can't be dug down to 26 symmetric clues at all. For 7 of the 200 expert puzzles, the closest the generator found was 27, and the book keeps that 27-clue puzzle rather than one that solves two ways.

The maze generator, measured on 2,000 mazes

The maze book generator page now has its own numbers. 500 mazes were generated at each grade, and every one was checked: every cell is reachable, and there is exactly one route from entrance to exit. One finding is useful to anyone making a graded book. Size alone does not set difficulty. The longest easy maze had a 179-cell route, longer than the shortest hard maze's 105. So before you publish, page through the preview. If an early maze looks too long for its place, Reshuffle remakes the whole book.

Criss-cross: a Medium book now says when a puzzle came out Easy

Measuring the criss-cross book generator turned up a fault. A puzzle that can't be built at the grade you asked for drops one grade rather than leave a gap in the book. That part is fine. But it happened silently. A Medium criss-cross book of West Virginia words came out as twenty Easy grids, with nothing printed to say so. Across 90 themes, 209 of 1,800 Medium puzzles had dropped a grade this way.

Now each of those puzzles is labelled with its real level, and the tool says how many there are before you download. You can paste in a longer word list, or knowingly accept the easier grade. Crossword books had the same fallback and got the same fix. The other half of the measurement was clean: all 1,800 grids, re-solved from scratch, had exactly one way to fill them.

Crosswords: how much of the grid is crossed

The crossword book generator page now has its numbers too: 7,200 puzzles, every answer with its clue. In these themed grids, the middle puzzle crosses only about one letter in seven with another answer. A dense grid crosses every square. So here the clues do the work, and every clue is written in plain language for that reason. The level warning added to criss-cross books this morning covers crossword books too. A third of Medium themes had been quietly giving some Easy grids.

Thanksgiving and Halloween tracing books: the submit-by days

The Thanksgiving tracing worksheets page now gives the dates for a Thanksgiving tracing book on KDP. KDP says low-content review is taking three to four weeks, so with that box ticked, submit by Thursday 29 October. Without it, KDP says a book goes live in 3 business days, so submit by Friday 20 November. For Halloween, a book with the low-content box ticked is too late this year. Without it, the Halloween tracing worksheets page says to submit by Tuesday 27 October. KDP's exact wording is quoted once, on the Christmas tracing worksheets page.

Halloween and Thanksgiving: the letters, from their own words

On the Halloween tracing worksheets, all seven tall letters turn up, but f only in leaf, so leaving leaf out would drop one. On the Thanksgiving tracing worksheets, feather is the only word with an f. Each page also names its quickest word to trace (bat at 4 strokes, mug at 3), its longest (9 strokes: pumpkin and lollipop; feather, pumpkin and seedling), and the words with an i, which take two strokes, the line and then the dot.

The Halloween book's cover, measured

The Halloween tracing worksheets open a 46-page book: A to Z, then twenty Halloween words. At 8.5" × 11" on white paper, its full-wrap cover is 17.35" wide by 11.25" tall, with a spine of about 0.10". KDP allows spine text only from 80 pages, so the Thanksgiving tracing worksheets book leaves its spine blank. Six words are on both lists: apple, acorn, candle, cookie, pumpkin and mushroom. An autumn book can carry one into the other. Every trim and page count is on tracing book cover size.

The margin table, in KDP's words

The margin calculator table now lists KDP's inside-margin ranges exactly as KDP's help page words them: 24 to 150 pages takes 0.375" (9.6 mm), 151 to 300 pages takes 0.5" (12.7 mm), 301 to 500 takes 0.625" (15.9 mm), 501 to 700 takes 0.75" (19.1 mm) and 701 to 828 takes 0.875" (22.3 mm). Outside margins are 0.25" (6.4 mm), or 0.375" (9.6 mm) with bleed. The source is linked and dated on the page, and the calculator still agrees with the PDF it makes in all 120 combinations tested.

Capitals take more strokes

The name tracing worksheet generator traces a name exactly as it is typed, so MAYA comes out in capitals and Maya as a capital and then lowercase. The numbered start dots, one for each time the pencil goes down, show what that changes. E takes 4 strokes, and A, F, H and I take 3 each. Every lowercase letter takes 1 or 2. Maya is 6 strokes and MAYA is 10; Christopher is 13 and CHRISTOPHER is 24.