2026-09-25 (fifth run): ten more common verbs, focus to gather
What this run did
This was the fifth scheduled run of the day. The mode selector (tools/next_mode.py, the script that picks each run's job from the project's state) picked build: writing new entries from the queue, the project's list of words still to be written. It chose build because building was the job furthest behind its share of runs.
Before starting, the run found no unfinished work: no open pull requests (proposed changes waiting to be merged), no leftover branches, and nothing in inbox/, the owner's instruction folder.
The run wrote ten verbs from the most common band of words (band 1): focus, fold, follow, force, forget, forgive, form, freeze, fry, and gather. The largest are follow (seven senses), form and freeze (six each). Force and gather also have a box on choosing between close words: force / make / compel (only make is followed by a verb without to), and gather / collect. Forget has a usage note on forget to do against forget doing, a real trap for learners.
Every entry went through the full checking pipeline:
- Word forms (forgot, froze, frozen...) came from the project's rules and its irregular-forms table; all confirmed.
- Pronunciation. All 20 American and British transcriptions were confirmed by a panel of other models and the CMU Pronouncing Dictionary, an open list of American pronunciations.
- Review. Two reviewer models from other companies read every field: reviewer-a (an OpenAI model) and reviewer-b (a Google model).
What the reviewers found and what was decided
The reviewers raised 39 issues, 25 of them blocking (the entry cannot be published until the issue is settled). The run read every issue and decided each one: 24 were applied, 15 rejected with a written reason, and none passed to the owner. The reviewers never write the entries; every fix was written by this run.
Real problems the reviewers caught, almost all reviewer-a:
- Two learner errors were wrong. The draft said He followed after me and I forgot my phone at home were mistakes. Both are normal English, so the errors, and the translator notes that repeated them, were replaced or removed. A fence line in
NEXT.mdnow records this. - The focus definition said a camera "sees" clearly. A camera shows a clear picture; the definition was split.
- Fold for a business said it closes "because there is not enough money". A newspaper or a show can also close simply because it fails, so the definition now says "because it is not successful".
- Gather speed was defined as "increase", which describes the speed, not the train. It now reads "slowly get more of something such as speed, strength, or support".
- Freeze wages was marked "usually passive", but the entry's own examples were active. The mark was removed.
Rejected issues included these:
- Reviewer-a again asked for a length mark in the American transcription of freeze. The project's settled rule is that American transcriptions drop it; the CMU dictionary agrees.
- Reviewer-a said The children formed a circle and Freeze! had the wrong grammar; both were right as written.
- Reviewer-b three times objected to the quoted headword at the start of translator notes, which the house style allows.
Reviewer-b's reply on form came back cut off; a re-run for that reviewer worked cleanly.
Five definitions used words outside the defining vocabulary (the plain word list definitions must use): lens, app, fail, arranged, respond. Each was reworded by hand.
Links and queue
Three existing entries gained a link back to a new one: crash (computer sense) to freeze, and boil and bake (cooking senses) to fry. All three were checked by hand and land on the right sense. While checking, the run noticed that boil sense 1 still begins its definition with "(of a liquid)", which the style guide moves to the explanation field; that is now item 8 in NEXT.md for the next review of that entry. Twenty-nine missing targets (focus the noun, freezer, compel...) were queued.
Spend
US$0.46 for this run (the review panel and pronunciation checks). US$1.86 so far today, of the US$5 daily cap.
What is next
The next build continues the band-1 verbs from give onward; give is a very large verb, so that run should stay near ten entries. A lint pass (the project's periodic whole-dictionary check) is due in about two more runs, and the originality check is due soon.
What needs the owner
Nothing new. The three open questions in wiki/open-questions.md are unchanged: whether to split be/have/do/one by part of speech, the look of the inline marks, and whether infinitive to gets an entry.