Until now, this dev journal existed only as notes on my own computer. This time, I made it readable on the SideKick Lab site. Both the Japanese and English versions went live — but getting there meant repeating the same mistake four times.
What went wrong
The SideKick Lab site uses individual static HTML files for each page. The shared header and footer are auto-generated from a template on some pages (such as the Uchide no Kozuchi page), but on many other pages the same header code is embedded individually, page by page.
When I added the navigation link for the dev journal, I assumed fixing the template side would be enough. It wasn't.
What I changed this time
The mechanism for generating and publishing the dev journal articles itself worked without problems. But when I actually looked at the site, there were pages where the nav link wasn't showing up.
There were two causes. One was that even after fixing the template, the change doesn't automatically carry over to already-generated static HTML files. The other was that eleven pages have the header code embedded individually, so template changes simply don't reach them in the first place.
I regenerated the affected pages and fixed the individually embedded pages one by one. Then the same problem turned up in the English version, because it had the same structure. In the end, between Japanese and English, I went through the same cycle of spotting and fixing the same kind of missed update four times over.
Partway through, there was also a small bug in how the article data was being loaded. The separator format in the article files was slightly different from what the template expected, which caused an extra line to appear at the top of the body text. I only noticed it once I actually ran real data through the system.
Why I took this approach
I could have solved this properly by templating the entire page structure, but I chose not to. There are 21 pages involved, and some of them are carefully built sales pages. Rather than touching everything at once right now, I decided to set that work aside for a calmer moment.
This time, I chose to fix each missed update as I found it and verify in the browser as I went. It's not a flashy approach, but given the current level of risk, I think it was a reasonable tradeoff.
What's still left
I've put off a full templating of the header and footer for now. To avoid making the same kind of oversight next time I do a similar fix, I've written down a checklist of steps to follow.
For the English translations, my current light-touch workflow is to have an AI do a first-pass translation, then read it back myself using Google Translate or similar to catch any obvious errors. A more rigorous process would be possible, but for something I update daily, that level of effort didn't seem practical.