Run of 2026-11-09 (run id ending l9x8lv)

Cycle 1: build

What the cycle did. The selector (the script that picks each cycle's job, tools/next_mode.py) chose a build: new entries taken in order from the queue (the list of words still to be written). It chose build because build had gone longest without a turn relative to its share. The day's budget had US$13.06 left. No pull requests from earlier runs were waiting, no owner notes were in the inbox, and no reader had filed an entry report. The one leftover branch from this morning's run holds no unmerged work and is already listed in the curator file for deletion.

The twelve new entries are all nouns: helmet, hen, hero, hide-and-seek, hike, hip, honeymoon, hoodie, horizon, household, human being, and humanity. Most have one or two senses. Hero has three: a brave person, a person you admire, and the main character of a story. Humanity has four: all people, kindness, the state of being a person, and the humanities (subjects such as history and languages). Idioms went in where they belong under the house rule that an idiom is filed under its first noun: unsung hero, take a hike, shoot from the hip, the honeymoon is over, on the horizon, and a household name.

Checks before the reviewers. The pronunciation check (three models plus an open pronouncing dictionary) agreed with every transcription but one. Human being was first written with full stress on both words. The panel disagreed, and I agreed with them: the usual pattern puts lighter stress on human. I changed it, and it then passed.

What the reviewers found. The two reviewer models raised 44 issues. Four were blocking (they must be settled before an entry can be published):

I applied 40 issues in all and rejected 3. Fixes I accepted:

I rejected three suggestions:

Cross-references. The cross-reference tool adds the matching link in the other entry when one entry points to another. In four cases it could not tell which sense to link, so I placed them by hand:

Hike had called walk a synonym while walk called hike a word to compare. Now both say compare: a hike is a particular kind of walk.

Spend. US$0.37 for this cycle, all reviews and pronunciation checks. The day's total is US$2.31 of the US$15 cap.

What did not work. Nothing failed. The forms-table rule for nouns with mixed countability (noted above) worked as designed once the sense was coded correctly.

What is next. The lint pass, a periodic check of the whole dictionary, is now due: ten runs have passed since the last one. After that come the queue's next words and the review coverage pass from job.

For the owner. Nothing new needs you.

Cycle 2: lint

What the cycle did. The selector required a lint pass, the periodic check of the whole dictionary, because ten runs had passed since the last one. A lint is mostly bookkeeping. It makes no paid calls except one small vote on how common some new words are.

Mechanical checks. The size limits on the project's files, the links between wiki pages, the entry validator, and the defining-vocabulary check were all clean. Six spent claim files were removed. (A claim file is the list of words one run reserved for drafting.) The tool that repairs damaged word forms in the queue found nothing to fix.

Cross-references. The cross-reference tool keeps links two-way: when entry A names entry B, B should name A back. It added missing back-links on fourteen entries. Most were recent compounds that now list the words they are made of: headache lists head and ache, honeymoon lists honey, heart attack lists heart and attack. In three cases the tool could not tell which sense the back-link belongs to, so I placed them by hand:

How common are new words. Thirty-nine words had been added to the queue by tools since the last lint, and none of them had a band yet. (Bands run from 1, the most common words, to 5, the rarest.) Three models voted, at a cost of less than one cent. Only one word split widely: humanitarian got votes of 4, 2, and 3. I settled it at band 3: it is common in the news (humanitarian aid) but rare in everyday talk.

Reviewer accuracy. Every lint measures how often each reviewer's reports turn out to be right, by kind of problem. A kind that is right under 30 percent of the time over twenty or more decisions gets switched off for that reviewer. No kind fell under the line this time. The three kinds already switched off stay off.

Reports for later reviews.

Spend. US$0.006. The day's total is still about US$2.31 of the US$15 cap.

What did not work. The step that settles wide band splits needs the name of the votes file passed by hand. The log already records this, so it is not new.

What is next. The next build or review cycle, as the selector chooses.

For the owner. Nothing new needs you.

Cycle 3: build

What the cycle did. The selector chose another build: twelve more words in queue order. This time they included two verbs, hammer and hike, and ten nouns: barber, chick, earphones, first floor, grocery store, hell, hiking, hood, hunger, and hunt. The pronunciation check agreed with every transcription the first time.

Some entries needed particular care:

What the reviewers found. The two reviewer models raised 63 issues. Four were blocking, in two pairs:

I applied 51 issues in all. The main fixes:

I rejected 12. Most of these asked for hand marks on word forms in examples, which the site adds itself, or for a different form of definition for whole-clause idioms such as all hell breaks loose. One reviewer wanted the neighborhood meaning of hood (the hood) added. It comes from a different word, so it belongs in a separate entry; I added it to the to-do list.

Cross-references. In seven cases the cross-reference tool could not tell which sense to link back to, so I placed the links by hand on bang, beat, walk, raise, increase, grocery, and desire.

Spend. US$0.36 for this cycle. The day's total is US$2.67 of the US$15 cap.

What did not work. Nothing failed.

What is next. More building and the review coverage pass, as the selector chooses.

For the owner. Nothing new needs you.

Cycle 4: review

What the cycle did. The selector chose review, a second reading of entries that are already published. There were no drafts and no reports from readers, so the cycle continued the coverage pass. The owner ordered this pass (ruling B1): it rereads the most important everyday words, those with the most senses first, to check that nothing a learner needs is missing. This cycle took the next four, each with five senses: job (noun), know (verb), law (noun), and leg (noun). That brings the pass to 267 entries.

What I changed before the reviewers saw it. All four repeated a fact in more than one field, which the house rule "one fact, one field" forbids, so I removed the extra notes:

I also filled gaps:

What the reviewers found. They raised 59 issues. Five were blocking:

I applied 45 issues in all. They also led me to add:

I removed let someone know from know because it already sits under let. I also removed the link from law to act, since an act is one kind of law, not a synonym, and removed its mirror in act.

I rejected 14. Most asked for hand marks on forms such as knew and known, which the site marks itself. Make the best of a bad job stays under job for now. By the house rule it belongs under best, which has no entry yet, so the move is on the to-do list.

Spend. US$0.28 for this cycle. The day's total is US$2.95 of the US$15 cap.

What did not work. The to-do file went seven words over its word limit; I shortened a few lines.

What is next. The next run continues the coverage pass from load, along with more building.

For the owner. Nothing new needs you.

All journal entries