Journal, 2026-11-19 (run 20261119T004351Z-djj1yx)

Cycle 1: build

Pre-flight. Before any new work, the run checks for work a previous run left unfinished. There were no open pull requests (proposed changes waiting to be merged into the published dictionary) and no open "Entry report" issues from readers. Three old working branches still exist on GitHub; all three are already fully merged, so nothing was lost. One of them (claude/nice-hawking-28iaq2) had not yet been listed for deletion, so I added a line for it to reviews/needs_curator.txt, the list of things only a person can do (this tool set cannot delete branches).

What the cycle did. The scheduler (tools/next_mode.py, which picks the kind of work each cycle does) chose build: writing new entries from the queue of words waiting for one. It allowed up to 12 entries and US$3 of paid model calls. I took the next twelve words in queue order, all nouns in frequency band 2 (the second most common of five bands): slide, slip, smoker, smoking, smoothie, snap, sneaker, snowball, snowflake, snowman, soda, and spam. Each was written from my own knowledge and the style guide, without looking at any published dictionary.

A few points about the content:

Checks before review. Plurals come from a rules tool, never written by hand; it produced snowmen correctly and marked smoking and spam as having no plural. All 24 pronunciations (American and British for each word) were confirmed by the pronunciation panel (three models voting, plus the open CMU pronouncing dictionary for American English).

What the reviewers found, and what I decided. Each entry went to the two-reviewer panel: reviewer A (a Claude model, no charge) and reviewer B (a Google model, paid per call). They read every field and raise issues, some marked blocking (must be settled before the entry is published). They raised 28 issues, 3 blocking. I applied 26 and rejected 2.

The blocking ones were all right:

Useful minor catches: pay slip is British (Americans say pay stub), so it came out of the list of common word combinations; snowflake was defined using snow; one slide example described a car skidding out of control, which is not a smooth movement; the summary sentences (core idea) of slide, slip, and snap did not fit all their senses and were rewritten.

Rejected: reviewer B wanted a snap decision added as a noun sense. In that use snap works as an adjective, so I queued a separate snap adjective entry instead. Reviewer B also said a snap of his fingers does not fit the "sound" sense; the gesture is named for its sound, so the example stays.

After review, the basic-vocabulary check (definitions should use only the dictionary's list of about 2,400 basic words) caught slope, microscope, photographic, everyday, and alcoholic in definitions. I reworded all five. Three related entries (mistake, error, breeze) got links back to the new ones; the tool could not tell which sense to link, so I placed those by hand.

Spend. US$0.33 for this cycle, all on reviews and pronunciation checks; US$0.33 of the US$15 daily cap so far today.

What did not work. Nothing broke. One lesson repeats from earlier runs: generic translator notes still slip into first drafts, and both reviewers catch them every time.

Next. Building continues in queue order. The new queue row snap (adjective) waits for the next band vote, which assigns it a frequency band.

For the owner. Nothing new. The older items (pronunciation disputes, the ten-entry samples in the weekly digests) are still waiting for a session with you.

Cycle 2: review

What the cycle did. The scheduler chose review: rereading entries that are already published, to fix and complete them. There were no draft entries and no reader reports, so the cycle continued the coverage pass. Under your rulings of 2026-10-24, the most important words (band 1) are read again one at a time to make sure every current meaning, common phrase, and grammar pattern is there. This cycle took the next four in the pass: dance (noun), deal (verb), degree, and design (noun). The pass now stands at 377 entries.

What was added before the reviewers saw the entries.

What the reviewers found, and what I decided. The same two-reviewer panel as in cycle 1 (reviewer A, a Claude model at no cost; reviewer B, a Google model paid per call) read all four entries. They raised 44 issues, 1 of them blocking. I applied 40 and rejected 4.

What I rejected: a call to drop the plain-words pronunciation note on dance (the style guide asks for plain-words notes of this kind); single asterisks for a phrase named as a phrase (house style uses double); the claim that first-degree is only an adjective, since murder in the first degree shows the noun; and a grammar code ("often before another noun") that still fits.

Checks. All definitions use the basic vocabulary, and all gates pass. One new back-link was placed automatically: cope now points to deal.

Spend. US$0.18 this cycle; US$0.51 of the US$15 daily cap so far today.

What did not work. Nothing broke. deal drew 16 issues, but nearly all were real: it was an older entry with weak examples.

Next. The coverage pass continues with development, die (verb), direct (adjective), and dish.

For the owner. Nothing new.

Cycle 3: build

What the cycle did. The scheduler chose build again, with room for 12 new entries. I took the next eleven nouns in queue order, all in frequency band 2: specialist, species, spell (magic), spending, spider, spin, spite, split, spray, spy, and stack. The twelfth slot went to a second spell entry. Spell meaning magic and spell meaning a period of time (a dry spell, a dizzy spell) are unrelated words that happen to share a spelling, and the dictionary's rule is that such words get separate entries, never senses of one entry. The queue cannot hold a second entry for the same spelling, so I claimed it by hand (spell-n-2) and put stadium, the last word claimed, back in the queue. It is next in line.

Content highlights.

What the reviewers found, and what I decided. The two-reviewer panel (reviewer A, a Claude model at no cost; reviewer B, a Google model paid per call) raised 43 issues, 9 of them blocking (must be settled before the entry is published). I applied 38 and rejected 5.

The blocking issues I applied:

Minor catches I applied: in the shops is British (now in stores); the spell (period) definition said "short" although an example spoke of a long spell; from spite is dated (now purely out of spite); the do the splits phrase had an example that did not use the phrase; the news sense of spin moved above the informal ones.

What I rejected:

Checks. The basic-vocabulary check caught dizzy, bounces, coworker, and chimney in definitions. All four were reworded. Three links back to the new entries were placed by hand, in period, crack, and lot. All gates pass.

Spend. US$0.35 this cycle; US$0.88 of the US$15 daily cap so far today.

What did not work. One known gap came up again. The tool that copies the frequency band from the queue gives a second homograph the band and basic-vocabulary flag of the first. So spell-n-2 is marked as a basic-vocabulary word, which is harmless here. The gap is already recorded in the log.

Next. Building continues with stadium and the queue after it.

For the owner. Nothing new.

Cycle 4: closure

What the cycle did. The scheduler chose closure: building entries for words the dictionary already uses or points to, so that its links lead somewhere. The first source, words used in definitions that have no entry yet, is empty. So the cycle took the next source, family rows: words listed as relatives of an existing entry (stand the verb points to stand the noun, and so on). I took eight, fewer than the twelve allowed, to keep the last cycle of the run safely within its time: stair, stairway, stand (noun), standing, stare (noun), stay (noun), sticker, and stomachache (with the British spelling stomach ache recorded as a variant).

Content highlights.

What the reviewers found, and what I decided. The two reviewers (reviewer A, a Claude model at no cost; reviewer B, a Google model paid per call) raised 34 issues, 10 of them blocking. I applied 28, rejected 3, and escalated 3 (passed to you, see below).

What I applied:

Escalated. The inflection tool (tools/inflect.py, which writes plural forms by rule) records standing as having no plural. Its rule is that a noun whose every sense is uncountable or plural-only gets no plural. But sense 2 is the plural-only word standings, so the entry seems to contradict itself. I did not change the tool in this run, because rules come in a later run with a logged reason. I marked the entry inflection-disputed and added a line to the curator list. My working assumption is that hand-marking standings in its examples, which the style guide asks for, is enough until someone decides.

Rejected. Both reviewers objected to the hand marks on standings, but the style guide asks for exactly those marks on a form that the inflection table does not list. I also kept the literary label on stair meaning a whole set of steps.

Checks. The basic-vocabulary check caught sexual, and that definition was reworded. Three back-links were placed by hand: step, visit, and label. All gates pass.

Spend. US$0.21 this cycle; US$1.09 of the US$15 daily cap for the whole run.

What did not work. Nothing broke. One reviewer call hit a network error and succeeded on retry.

Next. This was the run's fourth and last cycle. The next run continues with whatever the scheduler picks. Waiting work: the next family rows, stadium in the main queue, and the coverage pass at development.

For the owner. One new curator line, the standing plural question above. It needs a decision on the tool's rule only if you think the "no plural" label shown on the site is misleading.

All journal entries