2026-10-12, run at 12:44 UTC, cycle 3: the sixty-eighth lint pass
What this cycle did
This was a lint cycle: a maintenance pass over the whole dictionary and its working notes, with no new entries and no paid model calls. The scheduler (tools/next_mode.py, the script that picks each cycle's job) forced it because five runs had passed since the last lint, which is the limit.
The mechanical checks all passed:
- Caps. Every capped file is within its limit. The working wiki stands at 27,332 words of its 30,000. The style guide is at 3,465 of 3,500 words, and the routine instructions at 2,412 of 2,500.
- Links. All 520 relative links in the project's pages resolve.
- Vocabulary. No definition in any of the 2,215 entries uses a word outside the defining vocabulary (the list of about 2,400 plain words that definitions must be written in) unless that word has its own entry. One missing word was added to the queue.
- Claims. Two old claim files were pruned. These are the files a cycle writes to reserve words before drafting them, and both cycles that wrote them had already merged.
Back-links
The cross-reference tool (tools/crossref.py), run over every entry, adds "back-links". When entry A lists entry B as a synonym, antonym, or member of its word family, the tool adds the mirror link from B back to A. This time it added seven back-links in five entries. I checked each by hand, because the tool always attaches a back-link to the first sense of the target, and that is sometimes the wrong one.
Two needed fixing:
- The review in the previous cycle had removed come from the word family of income, because learners would not see the two as related. But come still listed income in its own word family, so the tool added the link straight back. I removed it from both sides. Lesson for later runs: a link removed from one side only comes back on the next full cross-reference pass. This is now noted in
NEXT.md. - personal lists public as an antonym for its sense "connected with someone's private life". The tool put the mirror on public sense 1 ("open to all people"). It belongs on sense 4 ("known about by many people", as in public life), so I moved it.
The other five (important, commercial, last, old, and past, pointing to the new adjectives powerful, private, and previous) are on the right senses.
Reviewer precision
The lint pass also measures how often each reviewer model's complaints turn out to be right, broken down by kind of complaint (a "family"). This is its precision: the share of its issues I accepted. A family that falls under 30 percent over at least twenty decisions is switched off for that reviewer. Nothing new crossed the line. Eight pairs stay switched off. reviewer-b's "spelling or format" family is at 28 percent, but the owner chose to keep it running. Overall precision is 72 percent for reviewer-a and 53 percent for reviewer-b. The figures are recorded in wiki/notes/reviewer-precision.md.
Other checks
The curator file (reviews/needs_curator.txt, the list of questions waiting for the owner) has no duplicate lines. Both questions in wiki/open-questions.md are still open and still accurate: whether continent names may appear in definitions, and whether the former and the following standing alone are nouns or pronouns.
Spend
None. Lint makes no paid calls. The day's total stays at US$9.39 of the US$15 daily cap.
What did not work
Nothing failed. The come/income back-link was a gap in the previous cycle's fix, not a tool fault. The tool did what it is designed to do.
What is next
The next lint is due in about five cycles. The build queue continues with band-1 adjectives after public. The next review starts with increase, industry, infection, and influence.
For the owner
Nothing new. The two open questions above still wait for a ruling whenever convenient.