Skip to content

Behind the Scenes: a training site moves, a Cloudflare redirect never fires, and I publish too soon

Edition 8 of the newsletter went out before the review that was supposed to clear it had finished. Not a typo sneaking past a reader, the whole review, run in the wrong order, on a piece with my name on it.

Which is fair warning for the rest of the week. Five tracks, four of them me rearranging my own furniture, and three separate moments where something went out ahead of its check. Last time it was thirty domains I had half-forgotten I owned. This week the thing that got past me was me.

The newsletter edition that beat its own review to the door

"AI Learning, Keeping It Human" Edition 8 is the research-heavy one: the two-year Khanmigo randomised trial, the Stanford SCALE reviews, Tutor CoPilot, Kestin, OECD, Gallup. Nine sources verified one at a time, an archive return-check, a cover image, an independent model check over the top. Carefully built, carefully sourced, and then published in the wrong order.

The reviews did run, just late. Seven findings from the voice gate, a verdict of "defensible with edits" from the evidence gate. Between them: a quote I had credited to the wrong researcher, a claim about Khanmigo's measured gain warmer than what the study authors actually wrote, and three places that needed a real evidence card rather than my confidence. All applied.

The part worth putting on the record is what happened to the sequencing mistake itself: nothing quiet. A retroactive review receipt, a partial-gate note in the edition's front matter, and a rubric-failure entry filed against my own process. The tempting version is the one where I fix the order and nobody ever knows. The filed version is uglier and considerably more useful in six months. Housekeeping closed out the same day: editions 6, 7 and 8 all marked published, K-12 tracked as a secondary audience, one stale delivery-evidence rule retired.

Moving the training pages onto their own domain in a day

Wednesday was Lakeview Training. A skeleton site (home, about, terms, privacy) onto Cloudflare Pages, then the move itself: twelve pages plus the archive off thisismyurl.com and onto the new domain, schema, sitemap and canonicals following. Then a full redesign on the far end, a ruul.io-inspired concept in lake ink, forest and sunrise gold, set in Fraunces and Public Sans. A visual check caught a content-security-policy rule quietly blocking the webfonts, plus a hover state nobody could read.

The one that cost me the afternoon was the redirect. The Cloudflare rule was configured, correct, sitting in the dashboard looking rather pleased with itself, and it never fired once. It is the shop that tapes a WE HAVE MOVED card in the window and then closes the blinds over it. Turns out this is a known gap on the Free plan: host-based rules fire, path-based rules just sit there being decorative. I pushed the redirect down to the origin in the timu-redirects mu-plugin and it worked on the first try, which was satisfying and slightly insulting.

Then the unglamorous middle: every internal link still pointing at the old /training/ path, five posts, four postmeta fields and one link-registry entry, one at a time.

A wordmark, a tee shop, an ad teardown, and twelve guides

Saturday was ShirtDock, a static tee storefront with Stripe checkout on Cloudflare Pages, which I revisited Thursday because it and Voidcharter had both shipped with no canonical tag, no Open Graph, no schema. Physician, heal thyself. Sunday was Lake America, compositing the wordmark onto four design directions. Wednesday also handed ditchtheguide.com a new design direction, deliberately over the top, now live.

Thursday was AdSectator, a new editorial site built from scratch with its first teardown on it, a look at the Novig and Sydney Sweeney campaign. Fact-check, voice review, value-editing pass, then publish, in that order this time. It is now a listed property on lakeviewbrands.com, with a monthly cloud routine keeping it running.

The pilot sites took the biggest block of hours. I pulled page-query data out of Search Console across the eight Niagara industry sites, found cannibalisation and some plain targeting mistakes, ran a full crawl and score over the top, and came out with an eighty-finding audit, the same site-audit shape I run on client search work. Then the fixes: a navigation fix, a cannibalisation fix, four content pieces, twelve new guide articles, four minimum per site, gate-reviewed twice. Twice, because the first pass caught a draft stating a hard technical claim it had invented outright. That one did not ship. I also corrected a wp-cli pull request description after a maintainer flagged it, checked against source and a second model before resubmitting, because by Thursday I had learned this lesson twice already.

What I took from the week

Three things tried to leave ahead of their check: a newsletter, a guide draft, and a pull request description. Two got caught by a gate. The one that did not is the only one of the three with my name at the top of it, landing in the inbox of somebody who asked for it.

The gates are not paperwork. I built them because at the end of a long build day I cannot read my own drafts honestly, and this week I proved it at my own expense. The bill arrived as a footnote in my own front matter, cheap next to the version where a wrong attribution sits in somebody's inbox and nobody writes it down. The receipt stays where I have to look at it.