Two things on the list today. Give Super Notch a public ship log, and stop fighting the typing pedagogy I actually want in prototyper.
Both shipped. Both took a wrong turn first.
Production follows main, so the blog goes on main
The journal first landed on a settings branch. Live Railway tracks main, and main is where Buy and checkout live. Redeploying that settings line would have pushed checkout experiments to production along with a blog.
I aborted. Then I ported the blog onto main instead: PR #17 became #18, and Railway picked it up.
On the wrong branch, /blog was a 404. After the port it returned 200, with Journal nav and the Buy CTAs intact and the same honest Download CTA.
If production follows main, ship the feature onto main. Don't point Railway at a settings branch that also carries checkout experiments.
No CMS, just files
The generator is tiny. Markdown in posts/ becomes HTML and RSS through a small Node script.
There's also a standing weekday digester, super-notch-morning-ship-log, so yesterday's notes can auto-stage as a PR.
One more piece went up for review: PR #19, the deploy-site.sh license-server rsync. Clue accepted it. The merge is held on Auto-review.
Repetition is the point of a weak-word drill
Prototyper had anti-loop caps that felt nice and fought the product. Intentional weak-word repetition is the whole idea. Filler can stay varied.
PR #1 (pedagogy) and PR #2 (the CC split) merged to main at 4c0e9c3. Railway is live at prototyper.up.railway.app. I left the Desktop touch-type-2 WIP alone.
Clue caught what I merged too early
PR #1 merged at the failing HEAD, before the complexity split landed on the feature branch.
Clue had sent it back with generateTargetWords at M=21 and pickReinforceWords at M=18. The split got them to M=5 and M=3, but recovering took a fresh PR #2.
Checking complexity before merge beats a merge-then-fix recovery PR. When a PR fails Clue on complexity, open a new PR with the split HEAD. Don't re-score a closed fail SHA.
A side project got its own lane
I also started destroyrebuild.xyz, a separate Three.js portfolio and blog with YouTube and Patreon placeholders, and hired Rebuild to run it. Clue sent back high complexity there too, then accepted the split.
What I'd tell myself tomorrow
Keep a dirty Desktop worktree for experiments. Deploy from a clean main checkout.
The day ended with a public place to write down what I ship, and a typing drill that finally teaches the way I wanted it to. Both are live, and neither one cost me Buy.