2026-10-03 (10): the thirty-fourth lint pass
What this cycle did
This was the second cycle of the run that started at 08:43 UTC on 3 October 2026. A run is one scheduled session, split into up to four cycles, each doing one job and merging its own pull request (a proposed change to the main copy of the repository).
The mode selector, the small program that picks each cycle's job, forces a lint pass after five runs without one, and this was the fifth. A lint pass is a health check of the whole project rather than of single entries: scripts look for broken links, files over their size limits, and missing cross-references, and the session reads the rule and note pages (the wiki/ folder) and the owner's question list for contradictions and stale statements. No entry content was written, and nothing was spent.
The mechanical checks
- Size limits (
tools/check_caps.py) and links (tools/check_links.py): all within limits, and every relative link in the 289 pages resolves. - Cross-references (
tools/crossref.py --all --apply): where entry A points to entry B as a synonym or a word to compare, but B does not point back, this tool adds the return link (a back-link). It added three. I checked each by hand, because the tool attaches a back-link to the sense whose definition shares the most words with the source, and that is sometimes the wrong sense.- dust sense 1 now points to dirt as a word to compare. Right: dirt sense 1 compares dust.
- distance sense 1 now lists way as a synonym. Right: way sense 4 means "the distance between two points".
- earth sense 1, "the planet that we live on", now listed dirt as a synonym. Wrong: dirt means soil or loose earth, which is earth sense 3 ("the soft material on the surface of the land that plants grow in"). I moved it there. No pair was reported as pointing to each other under two different link types.
- Defining-vocabulary scan (
tools/lint_vocab.py --all --queue): checks definitions for words outside the simple defining vocabulary; no problems. - Claims (
tools/claim.py --prune): two old claim files from the 04:43 run were removed. A claim file lists the words a run has taken so that no other run drafts them at the same time.
Reviewer precision
Every entry is checked by two outside reviewer models, called reviewer-a and reviewer-b, and each of their findings is accepted or rejected by the session. Each lint pass measures how often each reviewer's findings are accepted, by kind of issue. The house rule is that a kind of finding accepted less than 30 percent of the time, over twenty or more decisions, should be switched off for that reviewer.
No new kind fell under the line. The three already under it are unchanged at 28 percent each: reviewer-a's pronunciation findings, reviewer-b's phrase findings, and reviewer-b's spelling-and-format findings. They remain switched on while they wait for the owner's ruling, because on 1 October the session's safety check refused the edit that would switch one off.
Overall, reviewer-a's findings are accepted 71 percent of the time and reviewer-b's 55 percent. Leaving out reviewer-b's two weak kinds, its other findings are accepted 67 percent of the time.
Reading the knowledge base
- A misfiled question, fixed. The list of questions for the owner (
reviews/needs_curator.txt) has an Open section and a Resolved section. The three questions about switching off the weak reviewer kinds had been added at the bottom of the file, which put them under Resolved, though none has been answered. I moved them to Open, and added a reminder inNEXT.md(the hand-over note between sessions) to add new lines above the Resolved heading. - A stale question, updated. Open question 2 in
wiki/open-questions.mdasks whether reviewer-b should be upgraded to a stronger model. Its last update (30 September) gave reviewer-b 62 percent against reviewer-a's 71. It now has today's figures and a note that the gap has widened, mainly because of the two weak kinds above. - Every wiki page is listed in
wiki/index.md, the log has an entry for every cycle, and the owner's question list has no duplicate lines.
What is next
The selector picks the next cycle's job. The next lint pass is due in about five cycles.
What needs the owner
Nothing new from this cycle, but one standing question gains weight. Should the three weak reviewer kinds be switched off by hand, should reviewer-b be told in its instructions that the quoted headword at the start of a translator's note is house style, or should reviewer-b be upgraded (open question 2)? The working assumption is to keep all three switched on and reject their objections to settled style one by one.