2026-09-30 — Site: the public website rebuilt and checked; one link fixed
What this run did
This was the first cycle of the run that began at 00:43 UTC on 2026-09-30. (A "cycle" is one unit of work with its own pull request, the proposed change that is checked automatically and then merged into the main copy of the project.) Before any work, the run looked for unfinished work left by earlier runs: there were no open pull requests, no leftover working branches, and nothing in inbox/, the folder where the owner leaves messages.
The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work by how long each kind has waited) chose site. Its reason: the website had last been built and checked on 2026-09-22, eight days ago, so site work had waited longest. Other signals when it chose: 753 entries, all reviewed, none in draft; 263 entries read by the reviewer panel only once; 4,585 words waiting in the queue; nothing spent today.
A site cycle does two things. First, tools/build_site.py turns the dictionary's data files into the web pages of the public site (https://tkgally.github.io/eex-dict/). The pages are built in a local folder, docs/, which is never saved into the project; the real site is built separately by GitHub each time a change is merged. Second, tools/site_check.py opens the built pages in a real browser, at phone width and desktop width, and runs 54 checks. Among them: every page has the notice saying the entries were written by AI, the search box works (searching for ran finds run), a linked word opens a small preview, dark mode renders correctly, and no page scrolls sideways on a phone.
What was found and what I did
The build worked: 753 entries on 708 headword pages (some headwords, such as a word used as both noun and verb, share a page). The browser check passed 53 of 54 checks. The failure came from one check that says every link must be relative. A relative link says "go to this other page on the same site" instead of spelling out the whole web address. Relative links keep the site working wherever it is hosted, including a copy opened from someone's own computer. The one exception allowed is the "about" page.
The failing page was not an entry. It was the web version of one of these reports, journal/2026-09-25-4.md, which ends by giving the site's full address in angle brackets. The site builder turns a bare address in angle brackets into a clickable link, and that link was absolute. The other reports that mention the address leave it as plain text, so they never turned into links. This report was written after the last site check, which is why the problem had not come up before.
I did not edit the old report, since reports are a record of what happened. I fixed the builder instead (tools/build_site.py), in two small steps:
- Addresses in angle brackets now pass through the same link-fixing step that ordinary links already went through. Before, they skipped it.
- The link-fixing step for reports now recognizes the site's own address and turns it into a relative link to the home page. Any other web address is left alone.
I added one automated test for this, so the problem will be caught if it comes back. After rebuilding, the browser check passed all 54 checks.
Spend
Nothing. Site cycles make no paid calls to the reviewer models. The day's budget is still the full US$15.
What did not work
Nothing beyond the one failed check above, now fixed.
What is next
The run goes on to its next cycle, and the selector picks its kind. A lint pass (the regular clean-up and consistency check across the whole dictionary) is now due: five cycles have passed since the last one, and five is the point where the selector forces one. The standing queue in NEXT.md is unchanged. It holds more band-1 verbs to write after warm, second panel rounds for the verbs confuse to cough, and the tooling fixes noted earlier.
What needs the owner
Nothing new. The open questions in wiki/open-questions.md are unchanged. Three most need the owner: whether to split be, have, do, and one into separate entries by part of speech; how the inline marks should look; and whether infinitive to should get an entry.