Run report, 2026-11-17

Cycle 1: lint

Why this cycle was a lint. The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work) chose a lint pass because ten cycles had passed since the last one on 2026-11-15. A lint is the dictionary's housekeeping: it runs every mechanical check over the whole collection of 4,479 entries and then reads the project's own notes for contradictions and gaps. It writes no new entries, so its budget for paid model calls is nil apart from one small vote.

Before the work. No pull requests were left open by earlier runs. Two old working branches remain on GitHub; both were checked and hold nothing that is not already on main. One of them already had a line asking the owner to delete it; this run added the same line for the other (claude/nice-hawking-93pz5c), because the tools available to a run cannot delete branches. No entry reports (problem reports filed by readers through the site) were open, and the inbox was empty.

The mechanical checks. Word caps on the project's rule files and wiki were all within limits, and every relative link in the wiki resolves. The cross-reference tool (crossref.py), which makes sure that when entry A points to entry B, entry B points back to A, added four missing return links on its own: compliment-n now compares praise-n, roommate-n lists room-n in its word family, and rumor-n and rush-n each gained one. It left eight return links for a person to place, because it could not tell which sense of the target entry they belong to. I placed them by hand:

Six finished claim files (records of which words a past run had reserved) were cleared away.

The band vote. Fifty-two rows in the building queue had no frequency band yet (bands 1 to 5 rank how important a word is for a learner). Three language models voted on each, at a cost of under one US cent. Forty-nine were settled automatically; three had votes more than one band apart, and I settled them at the middle vote as earlier lints have done: attention as a spoken command (band 3), boy as an exclamation, as in Oh boy! (band 2), and rubbish as a British adjective meaning "bad" (band 3). All 52 are now ready to build.

Reports read. The list of possible missing senses (an entry links to another entry on only one sense when another sense may also fit) stands at 1,039 pairs; those are worked through in review cycles. The list of definitions that use their own headword stands at 314; most are harmless (for example "an illness caused by a virus" under virus). Six "promises" remain, where an entry's note says another word has its own entry and none exists yet; they are the same six as last time. Three (storey, the foot sense of corn, the tool sense of file) are named in the build plan; the other three are phrasal verbs or phrases (butter someone up, the double phrasal verbs, used to) that will be built in their turn.

Reviewer precision. Precision is the share of a reviewer's flags that I accepted. The first reviewer stands at 85 percent overall and the second at 83 percent. No new kind of flag fell below the 30 percent line that would switch it off.

Spend. About US$0.007, all on the band vote.

What did not work. The band-vote tool's settle and apply steps looked for a default vote file name and failed until I named the file explicitly; this was a small nuisance, not a fault in the data.

What is next. The next cycle goes back to the selector, most likely a build. The next lint is due after ten more cycles.

For the owner. Please delete the two old branches named in reviews/needs_curator.txt when convenient. Nothing else needs you from this cycle.

Cycle 2: review

What this cycle was. The selector chose a review cycle: a second reading of entries that already exist, in the order the owner set on 2026-10-24. There were no unfinished drafts and no open reader reports, so the cycle continued the coverage pass. The coverage pass rereads the most important everyday words (band 1, the top frequency band) and asks one question of each: does the entry cover every common present-day use, including idioms? This cycle took the next four words with four senses each: brother (noun), brown (adjective), build (verb), and button (noun). NEXT.md also listed lord-v (lord it over someone) for review, but that entry does not exist yet; it is a low-priority row waiting in the building queue, so I removed it from the review list.

What I added before the reviewers saw the entries. Brother gained two idioms, brothers in arms and Big Brother (a government or organization that watches everyone). Build gained the use for body shape (He was tall and strongly built), a pronunciation note that the u is silent so that build sounds like billed, and build a website among its collocations. Button gained the British idiom bright as a button (clever and lively). Brown gained brown paper.

What the reviewers found. Two reviewing models from different companies read every field of the four entries and raised 47 issues, 4 of them blocking (an issue that must be settled before an entry can count as reviewed). All four blocking issues were right, and I fixed each one:

Of the 43 minor issues I applied 33 and rejected 10. Among the changes: brother used to address a man (Thanks, brother) was defined as "a male friend", which both reviewers rightly said was wrong, so it is now defined as a form of address. One reviewer pointed out a common missing use, brother meaning a Black man (informal, American), which is now sense 5. The other pointed out Oh, brother!; that is an interjection, which is a separate part of speech, so I queued a new entry for it and pointed to it. The body-shape use of build became its own sense rather than a subsense, because it differs in both meaning and grammar. Button now uses the American fasten a button where it had the British do up. Several notes said the same thing twice and were trimmed. The rejections were mostly reviewer B asking four times to hand-mark the word built in examples. The site already does this for every form listed in the entry's table of forms, so hand marks would be wrong. I also rejected a request to add build bridges to build: it already lives under bridge, because idioms are kept under their first noun.

Spend. About US$0.17 on the reviewers, bringing the day to about US$0.17 of the US$15 daily cap.

What did not work. Nothing failed. One adjudication command was mistyped and logged a duplicate line, which I removed by hand before committing.

What is next. The coverage pass now stands at 357 entries. The next four are care (noun and verb), certain, and child.

For the owner. The new brother sense ("a Black man", informal, used mainly by Black people) and the rewritten brown sense about people are sensitive. Both were written with care, but you may want to look at them yourself.

Cycle 3: build

What this cycle was. The selector chose a build cycle: new entries drawn from the building queue in the agreed order. It allowed up to twelve. Because this was the third cycle of the run, and every entry needs room for the review pipeline (two reviewing models, then my reading of each issue), I took six and wrote them fully.

The six new entries.

Checks. The forms of each word came from the project's rules (tomatoes from its list of exceptions). Both American and British pronunciations of all six were confirmed by the pronunciation panel, a group of models that must agree, together with an open pronouncing dictionary.

What the reviewers found. The two reviewing models raised 31 issues, none blocking. I applied 24 and rejected 7. The main changes were these. The line of soldiers or trucks had been a subsense of the smoke-shape sense of column; it is a different meaning and is now its own sense. A good sailor / a bad sailor had sat under the sport sense, but it is about passengers, so it is now its own sense too. Ketchup had been listed as a word-family member of tomato, but it is not built from that word, so it moved to "see also". Diploma now covers the qualification as well as the paper. Sailor uses the American an avid sailor rather than the British a keen sailor. Two notes that repeated a fact stated elsewhere were cleared. I rejected two requests to change house-style formatting marks, one objection to a valid grammar code, one to articles in collocations, and one example complaint. The cross-reference tool could not tell which sense of degree should point back to diploma, so I placed that link by hand on the university sense.

A check that failed once. The final vocabulary gate caught the word steady in the definition of jogging. It is outside the list of words that definitions may use, so the definition now reads "the activity of running slowly for exercise, usually for some time".

Spend. About US$0.17 on pronunciation and review calls. The day stands at about US$0.34 of the US$15 cap.

What is next. The next build continues in queue order. There are 5,257 rows waiting to be built.

For the owner. Nothing new from this cycle.

All journal entries