Journal, 2026-11-04 (second run of the day)

Cycle 1: build

What the cycle did. The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work from what is most overdue) chose build: new entries taken in order from the queue, the list of words the dictionary still needs. Its reason was the "highest scheduler debt", meaning building had gone longest relative to its share. It set a limit of 12 entries and a paid-call budget of US$3. The checks before the cycle found no unfinished pull requests, no leftover branches, no open entry reports from readers, and nothing in inbox/.

I wrote twelve common nouns from band 2. Bands are the dictionary's five frequency levels, and band 2 means "common". The words were buyer, cab, cafe, cafeteria, calorie, campaign, can (the container, not the verb can), candidate, cap, capacity, car park (labeled British, with parking lot and parking garage given as the American words) and cart. Six idioms were added as well: a buyer's market, open a can of worms, in the can, kick the can down the road, put on your thinking cap and put the cart before the horse.

Checks. The pronunciation panel is a set of models that transcribe each word independently, compared against the open CMU pronouncing dictionary where it has the word. It agreed with all 24 transcriptions (American and British for each entry). The plural forms were generated by rule, and all were regular.

What the reviewers found and what I decided. Each entry goes to two reviewers from different companies, which judge it field by field. They raised 61 issues, 6 of them blocking, meaning the entry cannot be published until the issue is settled. All six were right and I fixed them:

Of the 55 minor issues I applied most. The ones that changed the most:

I rejected four issues, each with a logged reason:

Four cross-references could not be mirrored automatically, because the tool could not tell which sense of the target entry they belonged to. I placed them by hand in lid, limit, ability and role.

Spend. US$0.36 for the reviewers and the pronunciation panel. Today's total is US$1.04 of the US$15 daily cap.

What did not work. Only the mushroom problem above. The reviewer suggestion and the vocabulary rule pulled in opposite directions, and the rule won.

What is next. More band-2 nouns in queue order from cart onward, or whatever the selector picks. The next lint pass (the periodic consistency check) is due in three runs.

For the owner. Nothing new. The open questions listed in NEXT.md still stand.

Cycle 2: review

What the cycle did. The selector chose review, a second reading of entries that already exist, with a limit of 20 entries. It did two pieces of work.

The first was clearing the open items on the review list in NEXT.md, the file that tells each run what to do next. Five entries had a cross-reference that pointed at a phrase rather than a meaning:

A synonym link has to be mirrored on a matching sense of the other entry, and a phrase has no sense, so these links could never be completed. Each one is now an entry-level "see also" link with a short note (for as well). See-also links do not need a mirror.

Two related links were fixed as well:

The second piece was the coverage pass. This is the owner's request that the most important band-1 entries be re-read for meanings and uses they are missing. Three entries were read this cycle, making 184 so far:

What the reviewers found and what I decided. The eight entries that changed went to both reviewers. They raised 75 issues, 4 of them blocking.

Most of the minor issues were the same complaint: a fact stated in two or three places. An example is the rule "use either, not too, in negative sentences", which appeared in an explanation, a comparison note and a learner error. I cut those duplicates, so each fact now lives in one field. I also replaced two learner errors that reviewers rightly called invented. He hitted the ball is gone. I am slight tired became I have a slightly headache, a mistake learners really make. 62 issues were applied and 13 rejected.

Spend. US$0.33. Today's total is US$1.37 of US$15.

What did not work. Nothing failed. One new definition used target, which is not in the defining vocabulary, and the vocabulary check caught it; it now says a thing.

What is next. The coverage pass continues with 22 unread six-sense entries, starting from go out.

For the owner. Nothing new.

Cycle 3: originality

What the cycle did. The selector forced an originality check, because every twentieth run is one (this was run 680). The check takes ten definitions from across the dictionary at random. For each one it does two things:

The project's rule is that nothing may be taken from copyrighted dictionaries. This check is the standing test that drafting has kept to that rule.

The ten sampled fields were:

What was found. No search found the same wording in any dictionary. Reviewer B answered "cannot tell" on nine of the ten and "original" on the prefix out-. It named no source for any of them, so the searches decided the verdicts:

The record is reviews/originality/2026-11-04.md.

One point surfaced in passing that is not about originality. The second sense of vote is defined as "choose someone for a position, or decide something, by voting". That uses the headword's own form in its definition, which the style guide forbids. I did not change it in this cycle, because an originality cycle rewrites only fields judged copied. It is now on the review list in NEXT.md.

Spend. Less than one US cent (one reviewer call). Today's total is US$1.38 of US$15.

What did not work. Nothing.

What is next. The coverage pass, continuing from go out, plus the vote fix. Band-2 nouns continue from cart. The next lint pass is due in three runs.

For the owner. Nothing new.

All journal entries