Run report, 2026-11-07
Cycle 1: site
What this cycle did. The scheduled run starts each cycle by asking a selector (tools/next_mode.py) which kind of work is most overdue. This time it chose site: rebuild the published website from the dictionary's entry files and check it in a real browser. The reason it gave was "highest scheduler debt", meaning the site check had gone longest relative to how often it is meant to run; the last one was on 2026-11-05, and more than five hundred entries have been added since the one before that.
The build (tools/build_site.py) turned all 3,907 entries into 3,274 headword pages (a page holds every entry that shares a spelling, so race the contest and race the group of people sit on one page). No entries were held back as unfinished drafts, because there are none: every entry has been through the full review pipeline. The site is built into a folder that is never committed; GitHub's own build publishes it after this pull request merges.
The check (tools/site_check.py) drives a headless browser at a phone width (360 pixels) and a desktop width (1,280 pixels) and runs 79 tests: pages load with no JavaScript errors; no page scrolls sideways on a phone (tested on all 3,280 pages); searching for an inflected form such as ran finds run first; a misspelling such as abadnon offers abandon; phrases are found by their main words; hovering over a linked word opens a short preview; the "translator's view" toggle shows and hides the notes meant for translators and remembers its setting; dark mode renders properly; long entries get a sense menu; a cross-reference to a word with no entry yet is shown grey and unlinked, as your ruling C2 asked; and a mistyped address lands on a page that searches for what was typed. All 79 passed, so nothing in the site code needed fixing.
The weekly digest. The selector also reported that the weekly digest (your ruling E3) was due, so this cycle wrote journal/digest-2026-11-07.md. In short: 518 new entries and 129 re-reviewed entries over the week; the two reviewers were right 86 and 83 percent of the time, both up from 79; spend between US$2.92 and US$4.79 a day; nineteen calls made against both reviewers, almost all under standing house rules such as the keyword rule for idioms. The digest lists ten entries for you to sample.
Pre-flight. Before the unit, the run looks for work a previous run left unfinished. There were no open pull requests, no leftover branches besides main, no "Entry report" issues from readers, and nothing in inbox/.
Reviewers and decisions. None this cycle: the site mode writes no entries, so no reviewer was called and no decisions were logged.
Spend. US$0. The site mode makes no paid calls. The budget ledger rolled over at UTC midnight as designed: the previous day's US$2.98 moved into its history list.
What did not work. Nothing failed. One small oddity is worth recording for the next lint: the metrics show eight queue rows marked claimed although no branch other than main exists. The queue's sync command, run by every build and by the lint's claim pruning, resets such rows, so this cycle left them alone.
What is next. Further cycles in this run, if the clock allows, take whatever the selector picks next: most likely a build or closure cycle (new entries) or a review cycle (the coverage pass on band-1 entries, next from check-v). The next lint is due in five cycles.
What needs you. Nothing new. The standing items are the pronunciation disputes in reviews/needs_curator.txt and the ten-entry samples in the two latest digests, all left for a session with you under your ruling of 2026-11-05.
Cycle 2: originality
What this cycle did. Every twentieth run, the selector forces an originality check. It is a spot check that the dictionary's wording is its own and not borrowed from a published dictionary. This was run 720, so the check was due. The tool tools/originality_check.py picked ten fields at random from across the dictionary: definitions of tidy-v, strip-v, benefit-n, and appeal-n, the phrase definitions on the ground (ground-n), pull your socks up (sock-n), and put your feet up (foot-n), and the usage notes of daily-adv, dress-n, and seeing-conj. For each, the run searched the web for the exact wording to see whether any published dictionary uses it. It also asked one outside reviewer model (reviewer B) whether each text reads as copied. The reviewer has to quote the source if it says "copied".
What was found. Seven came out original: no published match. Two came out generic overlap: short, plain wording that any dictionary would arrive at independently. These were remove a covering or a layer from something (strip-v) and sit or lie down and rest (put your feet up). One came out rewrite: the definition of pull your socks up, "start working harder or behaving better, because your work or behavior has not been good enough". A well-known learner's dictionary defines the idiom as making an effort to improve your work or behaviour "because it is not good enough". The idea is the idiom's own, but the matching clause and the pairing of work or behavior were close enough that I took the cautious verdict. Reviewer B called all ten original or "cannot tell" and named no source. It did remark that the socks wording was structurally like published definitions. As in past checks, the web search decided.
The rewrite. The rules treat a rewrite as a full review of the entry. The new wording is "try harder and do better after a time of doing badly or not making enough effort". The entry then went through the whole pipeline: format check, inflections, pronunciation check (both transcriptions verified), and two passes of the two-reviewer panel, because the first pass led to new material that also needed reading. Neither reviewer objected to the new definition. They raised seven minor points. I applied five: the main definition now says a sock covers "your foot and ankle, and sometimes part of your leg", since ankle socks reach no higher. The idioms put a sock in it (informal, British: stop talking) and work your socks off (informal: work extremely hard) were added. Wool socks moved from the adjective-plus-noun collocations to noun-plus-noun. The new put a sock in it got a plain verb definition, "stop talking or making noise", with its rudeness explained separately. I rejected two of reviewer B's label suggestions. It wanted put a sock in it not marked British, though it is chiefly British. It wanted work your socks off marked British, though it is common in both varieties. Nothing was escalated. The entry is back to reviewed.
Spend. About US$0.06 (US$0.005 for the originality question, the rest for the two panel passes).
What did not work. Nothing failed.
What is next. The next cycle takes whatever the selector picks. The next originality check is due in twenty runs, the next lint in four.
What needs you. Nothing new from this cycle.
Cycle 3: build
What this cycle did. The selector chose build: write new entries from the queue (the ordered list of words still to be written), at most twelve in a cycle. Twelve band-2 nouns were claimed in queue order. (Band 2 is the second of five editorial frequency bands; band 1 holds the commonest words.) One of the twelve, facts, turned out to be just the plural of fact, which already has an entry. A cross-reference in the entry data-n had queued it by mistake. It was marked a duplicate, and fixing that link is listed for data-n's next review. The other eleven were drafted from scratch: expression, extension, extent, factor, fame, fare, farming, faucet, favorite, feel, and female, all as nouns.
Some highlights. extent-n has a single sense, but the word is mostly used in fixed phrases, so it carries five: to some extent, to a large extent, to what extent, to a lesser extent, and to the extent that. It also has a learner-error note on the common slip in some extent. favorite-n covers the thing you like best, the competitor expected to win, a person treated better than others, and saved websites, plus the American idiom play favorites. feel-n separates how something feels to the touch (the feel of cotton) from the character of a place (a relaxed feel), with the idioms get the feel of something and have a feel for something. female-n has a usage note: as a noun for people, female belongs to science and official reports and can sound cold or rude in conversation.
What the reviewers found and what I decided. Each entry went through the full pipeline: the format check, generated plural forms, the pronunciation check, and the two-reviewer panel. Five entries changed enough after the first panel pass that the panel read them a second time. In all, the reviewers raised 78 issues, 11 of them blocking (an issue that must be settled before an entry can be published). I applied 60 and rejected 18; none needed escalating to you. The important fixes:
extension-nlacked the computing senses: the letters at the end of a file name (.txt), and a small program added to a web browser. Both were added. So was a subsense for the figurative an extension of her personality.factor-nhad the sun-cream sense (factor 30) marked uncountable even though its examples said a high factor. That was corrected. The second pass caught that the maths examples stated checkable facts (Three is a factor of twelve), which house rules forbid, so they became classroom tasks (Find all the factors of thirty). A subsense for the wow factor and the phrase the X factor were also added.fare-nhad tucked the entertainment use (the usual holiday fare) into an explanation. It now has its own subsense, and the food sense moved ahead of the taxi-passenger sense.favorite-nlisted a favorite among someone, which is ungrammatical. It became a favorite among students.
The rejections were mostly one repeated objection: that an ordinary noun should never have an explanation field. House style puts a restriction on use there (usually with an adjective, only in the phrase by a factor of), so I kept those. Claim to fame was asked for under fame-n, but the keyword rule (an idiom goes under its first noun) puts it under claim-n.
All 22 transcriptions (American and British for each entry) were verified by the pronunciation panel and the open CMU data. Three cross-references needed a matching link placed by hand in the other entry, because the tool could not tell which sense to attach it to: equation-n, addition-n, and ticket-n.
A small finding. Cycle 1 mentioned eight queue rows marked claimed with no branch behind them. The queue's sync step in this build marked them done, since their entries already exist. Nothing is left over.
Spend. US$0.53 for this cycle, US$0.59 so far today of the US$15 daily cap.
What did not work. Nothing failed. A shell loop that logged decisions dropped the rows that had an empty column, because Bash merges empty tab-separated fields. Those decisions were re-logged from Python. Nothing was lost.
What is next. The next build continues in queue order after female-n. The next lint is due in three cycles.
What needs you. Nothing new from this cycle.
Cycle 4: review
What this cycle did. The selector chose review: reading existing entries again rather than writing new ones. There are no unreviewed drafts and no open "Entry report" issues from readers, so the cycle continued the coverage pass you asked for in ruling E2. That pass reads the band-1 entries (the commonest words) with five or more senses, looking for senses and idioms that are missing. It also brings the notes into line with the "one fact, one field" rule. The next four in the pass were check-v, cold-adj, color-n, and comfortable-adj.
What was added before the reviewers saw it.
check-v: a subsense for check yourself (stop yourself suddenly: He started to speak, then checked himself) and a sense for attacking the king in chess. Chess is not in the defining vocabulary and has no entry yet, so the definition says "in a board game" and the explanation names chess.cold-adj: a subsense for cold logic and the cold truth (based on facts, without feeling); the guessing-game sense (you're getting colder!); cold colors (cold blues and grays); and the idiom out cold (unconscious).color-n: a sense for paint or dye (a natural color on her hair) and the phrase off color. The phrase means slightly rude in American English (off-color jokes) and slightly ill in British English.comfortable-adj: a sense for a comfortable pace and comfortable walking distance.
Notes that said the same thing twice. All four entries repeated facts across fields. The worst was color-n: the warning that colored is offensive appeared four times. One copy also said a person of color is used "across English-speaking countries", contradicting the American label beside it. Each fact now lives in one place. Other cuts:
comfortable-adjdescribed its three-syllable pronunciation in two fields. It also had a filler learner error (This chair is very comfortably), of the kind your notes rules exclude.check-vrepeated its control versus check false friend in both a learner error and a translator note.
What the reviewers found and what I decided. 41 issues, 2 blocking. Both blocking issues were circular definitions in senses I had just added, where the headword appeared in its own definition. Cold colors became "making you think of ice, snow, or winter". Paint or dye became a substance "that makes something red, blue, green, and so on". I applied 34 and rejected 7. Among the applied fixes:
- A cold drink is chilled, not "allowed to cool", so the food sense now covers drinks.
- Food color became a different example, since Americans say food coloring.
- Comfortable income fits a widened definition: "having or giving enough money".
The rejections: leave someone cold already lives under leave-v by the keyword rule. Colder needs no hand marking, because the site marks listed forms itself. A cold, hungry child shows that the feeling-cold sense can come before a noun. Nothing was escalated to you.
One pronunciation question. The pronunciation check runs a panel of three models for British English. This time it gave the existing British transcription of comfortable-adj (ˈkʌm.fə.tə.bəl) only 1 vote in 3; in October it got 2 in 3. The transcription itself did not change. There are no open British pronunciation data to settle it, so the entry now carries the pronunciation-disputed flag. A line in reviews/needs_curator.txt records the working assumption: keep the careful four-part form, with the common three-part ˈkʌmf.tə.bəl as a variant. It joins the other pronunciation disputes left for a session with you.
Spend. US$0.22 for this cycle. The run as a whole spent US$0.82 of the US$15 daily cap.
What did not work. Nothing failed.
What is next. This was the run's fourth cycle, the most a run may hold, so the run ends here. The coverage pass continues with company-n, connect-v, continue-v, and country-n. The next lint is due in two cycles.
What needs you. Only the new pronunciation line above, alongside the standing items listed in this week's digest.