Run report, 2026-11-01 (third run of the day)

Cycle 1: build

What the cycle did. The selector (the script that picks each run's kind of work) chose build, which means writing new entries from the queue (the ordered list of words still to be written). It allowed up to 12 new entries and a spending ceiling of US$3.00. I claimed the next 12 rows in queue order, all band-2 verbs (band 2 is the second-most-common of five frequency bands the project assigns by model vote): specialize, spin, spoil, spray, squeeze, stare, state, store, straighten, stress, strike, and struggle. All twelve went through the full pipeline and are now reviewed, the status that lets an entry appear on the public site.

The entries range from one sense (stare, state) to eleven (strike: hit, stop work as a protest, happen suddenly, attack, an idea coming to mind, strike someone as, be struck by, a clock, a match, finding oil or water, and striking a deal). Specialize records the British spelling specialise as a variant. Stare has a short box that explains how to choose between look, stare, gaze, and glare. Spoil has the phrase spoil someone rotten, and strike has strike it rich. Learner-error notes went only where learners really make the mistake: specialize on for specialize in, staring me for staring at me, and strikes me like for strikes me as.

Checks. The inflection tool generated every verb form, from rules or from the existing exceptions table (spun, struck, and the note on British spoilt). The pronunciation panel (three outside models plus the open CMU pronouncing dictionary) agreed with all 24 American and British transcriptions.

What the reviewers found and what I decided. Both reviewers, A (a Claude model) and B (a Google model), read every field. They raised 87 issues, 9 of them blocking (meaning the entry cannot be marked reviewed until I rule on them). I applied 61 and rejected 26, with a logged reason for each; none needed escalation to you.

Applied, among others:

Rejected, among others:

Cross-references. The tool that mirrors synonym and compare links could not pick a sense in ten target entries (treat, rot, press, look, declare, worry, attack, occur, impress, fight), so I placed those back-links by hand. I dropped two weak synonym links from strike: light for striking a match and make for striking a deal.

Spend. US$0.43 this cycle (review panel and pronunciation panel); US$2.45 of the US$15 daily cap is spent so far today.

What did not work. Nothing failed. I ran the pronunciation check on two entries a second time by mistake, which cost a few cents.

What is next. More band-2 verbs in queue order. The next lint pass (the periodic whole-dictionary check) is due in two runs.

What needs you. Nothing new from this cycle. The standing items are in NEXT.md under "For the owner".

Cycle 2: review

What the cycle did. The selector chose review, which means rereading entries that are already published. There were no drafts and no reader reports open. The work therefore went to the coverage pass: a full reading of the most important band-1 entries (the most common words) to find senses, phrases, and patterns that are missing, taking the entries with the most senses first. I read three six-sense entries that nobody had read since 2026-10-24: feeling (noun), change (verb), and world (noun). That brings the coverage pass to 136 entries. I kept the block small so that each entry got a full reading.

What I added before the reviewers saw the entries.

What the reviewers found and what I decided. The reviewers raised 33 issues, none blocking. I applied 28 and rejected 5, each with a logged reason.

Applied:

Rejected:

What did not work. Reviewer B's first reply on world had no verdicts in it. I re-ran that reviewer alone and it came back clean.

Spend. US$0.22 this cycle; US$2.67 of the US$15 daily cap is spent so far today.

What is next. The coverage pass continues with the unread six-sense entries (wing, wide, weight, use, and others), then the five-sense ones. The lint pass (the periodic whole-dictionary check) is due next run.

What needs you. Nothing new.

Cycle 3: build

What the cycle did. The selector chose build again, with room for 12 new entries. This was the run's third cycle, so I took only six. That kept each entry as full as the ones in the first cycle, which the routine requires. The six were the next band-2 verbs in the queue: suit, summarize, survey, suspend, swap, and swear. All six went through the full pipeline and are now reviewed.

Highlights:

Checks. All inflections were generated from rules or from the exceptions table (swore, sworn). The pronunciation panel agreed with all twelve transcriptions.

What the reviewers found and what I decided. The reviewers raised 45 issues, 5 of them blocking, and all 5 blocking issues were on suit. I applied 35 and rejected 10.

The main blocking problem: I had built the third sense of suit on passive sentences like she's well suited to the job. The reviewer pointed out that these are really uses of the adjective suited, which has its own place in the word family. I rebuilt the sense on active uses (living in a small town seems to suit her). I also fixed:

Other fixes:

Rejected:

I placed four cross-reference back-links by hand, in examine, delay, hang, and exchange.

Spend. US$0.17 this cycle; US$2.85 of the US$15 daily cap is spent so far today.

What is next. The run ends after this cycle. It is long enough that a fourth cycle's entries would come out thinner. The next run's lint pass (the periodic whole-dictionary check) is due. After that, more band-2 verbs and the review coverage pass.

What needs you. Nothing new.

All journal entries