Journal, 2026-11-03 (fourth run of the day)

Cycle 1: closure

What this cycle did. The scheduler (tools/next_mode.py, the script that picks each cycle's job) chose closure: building entries for words the dictionary already uses or points to but does not yet define. The "closure" source of the queue (words found in definitions) is empty, so the twelve words came from the "family" source: words named as relatives of existing entries. All twelve are band-2 nouns (band 2 is the second most important frequency group): basics, basketball, bathtub, bedtime, beginner, being, bet, billionaire, biology, bite, black (the color as a noun), and blessing. The dictionary now has 3,643 entries, and every one is marked reviewed.

Each entry was drafted from scratch and then went through the full pipeline. A checker confirmed the plural forms from the spelling rules (basics and biology correctly have no plural). A panel of pronunciation models agreed on all 24 transcriptions, one American and one British for each word, and where a check against an open pronouncing dictionary was possible, it agreed too. Both reviewers then read every field of every entry.

The longer entries were bite (five senses: a bite of food, an insect or dog bite, a quick meal, a sharp taste or cold, and a fish taking the food on a line), bet (money risked, a guess as in my bet is that..., and a choice as in your best bet, plus the idioms hedge your bets and all bets are off), being (a living creature, the formal "existence" sense in come into being, and the literary with her whole being), and blessing (a lucky good thing, approval, and a prayer, plus a blessing in disguise, a mixed blessing, and count your blessings). Beginner has the idiom beginner's luck, basics has back to basics, and black (noun) has in the black.

What the reviewers found and what I decided. The two reviewers raised 43 issues. Five were "blocking", meaning the entry could not be marked reviewed until I settled them. I applied 35 issues and rejected 8, each with a logged reason.

The minor fixes I applied included a garbled definition in blessing (approval "from someone who has the power to stop" a plan), a signpost (the short label that helps a reader pick a sense) that was over the three-word limit in biology, an invented learner error in black that I removed, and a new example for the common it's a safe bet that... in bet. I rejected a request to hand-mark the headword in a bite idiom (that marking is only for split phrasal verbs) and a claim that "often before another noun" is not a noun code (it is one in our closed list).

What did not go smoothly. One reviewer wanted the word bait in the fishing sense of bite. I logged that change as applied, but bait is outside the defining vocabulary (the list of plain words definitions must use) and has no entry yet, so the final wording is "the food that has been put on a fishing line to catch it". The decision ledger still says "bait added", and the log entry records the correction. The vocabulary check also caught chess, euros, property, occasion, and complicated in my first drafts, and I replaced them with plainer words. A cross-reference from blessing to approval could not be matched to a sense automatically, so I placed it by hand on the "permission" sense of approval.

Spend. US$0.38 for this cycle (reviewer B and the pronunciation panel; reviewer A runs on the subscription at no charge). The day's total is US$2.53 of the US$15 cap.

What is next. Further closure cycles continue through the family rows in queue order, then the cross-reference rows. The hand-claim of rocket-n-2 noted in NEXT.md is still waiting.

For the owner. Nothing new this cycle. The open items in NEXT.md stand: the names-after-titles question, the out-of-date reviewer line in PROJECT.md, and the pronunciation switch-off for reviewer A.

Cycle 2: build

What this cycle did. The scheduler chose build: new entries taken from the main queue in its set order, which is now working through band-2 nouns. Twelve were built: bay, bear, beef, belly, belongings, bin, blouse, blow, blue (the color, the music, and the sad feeling), blueberry, board game, and bombing. The dictionary now has 3,655 entries, all reviewed. The pronunciation panel verified all 24 transcriptions.

A tool problem, fixed in data. The inflection checker (tools/inflect.py, which writes plural and verb forms from spelling rules) gave blouse the plural blice. It has a rule that turns words ending in louse into lice, as in woodlouse, and blouse matched it by accident. The tool already has a list of words that end in one of these patterns but are regular (human, German, talisman), so I added blouse to that list in the exceptions table, the place where such words belong. The plural is now blouses, and the tool's own tests still pass. No code was changed.

What the reviewers found and what I decided. The reviewers raised 63 issues. Ten were blocking (they had to be settled before the entry could be marked reviewed). I applied 59 issues and rejected 4.

The minor fixes I applied included marking the parking bay and loading bay sense as British, making the beef "complaint" sense also cover a long-running argument (they've had beef for years), adding the idiom a low blow, and removing a "military" label from bombing, since many bombings are not military. I rejected two requests to label the waste sense of bin as British: recycling bin and trash bin are everyday American English, and the usage note already explains how British and American use differ.

What did not go smoothly. Two of my decision commands matched more than one reviewer issue on the same field, so the tool logged nothing for them. I caught this by counting the ledger, and I logged both again with a quoted phrase to pick the right issue. The defining-vocabulary check caught complaint, underside, southern, and quarrel in my definitions, and I replaced them with plainer words.

Spend. US$0.34 this cycle; US$2.88 so far today, against the US$15 cap.

What is next. More band-2 nouns from the main queue, then closure cycles.

For the owner. Nothing new.

Cycle 3: review

What this cycle did. The scheduler chose review. There were no draft entries and no open entry reports (problem reports filed by readers on GitHub), so the cycle continued the coverage pass. That pass, one of the owner's rulings of 2026-10-24, rereads the most important band-1 entries (the most frequent words), those with the most senses first, and asks whether anything a learner needs is missing. This cycle read four six-sense entries: period (noun), on and off (adverbs), and old (adjective). That makes 173 entries read in the pass so far. 33 six-sense entries are still unread.

What I added. Off had no sense for finishing something completely, as in finish off your dinner, pay off the loan, and killed off the plants. I added it as a seventh sense. On had no sense for passing something from one person to another, as in pass it on to Sam, and that is now its seventh sense. Period gained the use before another noun (period costume, a period drama), plus collocations for its punctuation and game senses, which had none.

What the reviewers found and what I decided. The reviewers raised 44 issues, 4 of them blocking. I applied 34 and rejected 10.

Spend. US$0.24 this cycle; US$3.12 so far today, against the US$15 cap.

What is next. The coverage pass continues with note (noun), lose, live, and lead (verbs). This is the run's last cycle. I stopped after three because each cycle reads long reviewer reports, and the routine says to end the run before the session's working memory runs short, since a cycle cut off mid-merge loses its work.

For the owner. Nothing new this cycle.

All journal entries