2026-09-27 (2) — Lint: the eighteenth housekeeping pass
What this run did
This was the second cycle of the scheduled run that began at 18:44 UTC. A "cycle" is one unit of work that is checked, recorded, and merged on its own. The mode selector (tools/next_mode.py) forced a lint pass because five runs had gone by since the last one. A lint pass is housekeeping. Scripts check the whole dictionary and its notes. Then I read their output and the project's own pages for things a script cannot judge.
What was checked, and what it found
- Word caps and links. Every page is within its word limit, and every relative link resolves.
- Cross-references. When entry A lists entry B as a synonym or a word to compare, B should list A back.
tools/crossref.pyadds those mirror links, but it has to guess which sense of B to attach them to, and it guesses wrong about one time in five. So every lint pass checks each one by hand. This pass it wrote six. Five were right: deliver/send, divide/share, fix/set, govern/rule, and lay/set. The sixth attached refer (as in refer a patient to a specialist) to the mail sense of send. I moved it to the sense make someone go, where it belongs. - Vocabulary queue. A second script finds words used in definitions that have no entry yet and adds them to the list of words to write. It found nothing new.
- Stale claims. Two old claim files were removed. A claim file is how a run reserves headwords so that two runs do not write the same word.
- Reviewer precision. Each finding from the two reviewer models is either applied or rejected. Precision is the share that gets applied. For each kind of finding, any reviewer below 30 percent over twenty or more decisions stops being asked about that kind. No new kind crossed the line this time; five were switched off earlier and stay off. Two kinds are close:
- The first reviewer's pronunciation findings: 32 percent over 112 decisions.
- Its inflection findings (the forms of a word, such as ran and run): 6 percent, but over only 16 decisions, four short of the twenty needed. The figures are in
wiki/notes/reviewer-precision.md.
- The curator list.
reviews/needs_curator.txtis the list of items for you. Two lines asked for old branches to be deleted. Both branches are already gone, so I closed those lines. - Index, open questions, decisions. The wiki index lists every page. The open questions still carry their working assumptions, and nothing contradicts the decision pages.
One gap found
Your commit of 2026-09-27 raised per_run_cap_usd in config/routine-config.json from US$1.25 to US$3.00. That setting is the spending ceiling for one cycle. Every change is meant to have a line in wiki/log.md, and this one had none. I have recorded it in this pass's log entry. Nothing else needed to change: config/budget.md describes the setting without naming a figure.
Spend
US$0. A lint pass makes no paid calls. The day's total is still US$0.56 of the US$15 cap.
What did not work
Nothing failed.
What is next
The next lint pass is due in five runs. The selector still owes many review cycles, and the next ones should continue with the thinner verbs written late on 2026-09-26 (see NEXT.md).
What needs the owner
Nothing new. Open questions 4, 5 and 6 in wiki/open-questions.md are still waiting for you.