2026-10-04 (4) — The thirty-seventh lint pass
Fourth and last cycle of the run that started 00:43 UTC. A "cycle" is one unit of work that ends in its own merged pull request.
What this cycle did
The mode selector forced a lint pass because five cycles had passed since the last one. A lint pass checks the whole dictionary and its knowledge base (the wiki/ folder of rules and notes) for problems that build up over time. It runs mechanical checks with the project's tools and then reads things a script cannot judge. No paid model calls are made.
Mechanical checks. The size caps on the rule files, the links inside the wiki, the defining-vocabulary check (definitions must use a list of about 2,400 plain words, or words that have their own entry), and the claim files all passed. claim.py --prune removed four old claim files whose words now all have entries. A claim file reserves words for one run so that two runs never draft the same word.
Back-links. The cross-reference tool keeps links symmetric. If entry A lists entry B as a synonym, B should list A too. When it adds the missing half (a "back-link"), it attaches it to the target's first sense, which is sometimes the wrong sense, so every lint pass checks each one by hand. This pass found four:
- head (in its sense "the top or front end of something") lists front as a synonym. The tool had put the mirror on front's first sense, "the forward side" (the front of a house). The right home is the second sense, "the forward position" (the head of the line means the front of the line). I moved it.
- head in the same sense lists foot as an antonym (the head and the foot of a bed). The tool had put the mirror on foot's body-part sense. I moved it to the "bottom part" sense, where the foot of the bed is.
- ceiling as the antonym of floor, and house (a company, as in a publishing house) as a synonym of firm, were in the right places.
The tool also reported one type mismatch: folder named file as a synonym, while file named folder as a "compare" (a word learners confuse with this one). Since the folder usage note explains how the two overlap and differ, "compare" is the better type, and both sides now agree.
Reviewer precision. Each lint pass measures how often the session accepts each reviewer's objections, by kind of objection. The house rule says that a kind of objection accepted less than 30 percent of the time, over twenty or more decisions, should be switched off for that reviewer. No new pair fell under the line. The two pairs already under it are still held for your ruling, as before:
- reviewer-a on pronunciation is at 27 percent (53 accepted out of 199). Nearly all of its rejected objections are the same settled point: American transcriptions here leave out the length mark. This run rejected three more of these, on food and beach.
- reviewer-b on spelling and format is at 29 percent (142 out of 498).
Reading. I read the wiki index against the pages (every page is listed), the open questions (none is stale), and the curator file, which holds questions only you can settle (no duplicates).
What changed
Five entries: floor and firm gained their back-links; on foot and front the back-links were moved by hand; folder changed its cross-reference type. Also changed: the reviewer-precision note, the log, NEXT.md, and the queue (the vocabulary check queued words that definitions use but that have no entry yet).
Spend
Nothing; lint makes no paid calls. Today's total stays at US$1.41 of the US$15 daily cap.
What did not work
Nothing failed.
What is next
The next run's selector will most likely choose build or review. The next lint is due in about five cycles.
What needs the owner
The same standing items: whether to switch off the two low-precision reviewer pairs by hand (an earlier session's safety check refused the edit), and open questions 4 to 7 in wiki/open-questions.md, especially 7 (proper names in definitions), which keeps Christian out of the dictionary.