Run report, 2026-10-27 (fourth run of the day)

Cycle 1: build

What the cycle did. The selector (the tool that picks each cycle's job from what is most overdue) chose a build cycle: writing new entries. It built ten, each through the full checking pipeline, and all ten are now published.

Five of them are second homographs: words spelled like an existing entry with the same part of speech but an unrelated meaning, so they get an entry of their own with -2 in its name. They had been waiting for a hand claim because the queue (the list of words still to write) has no way to hold a second homograph:

The other five are adjectives named in the word-family boxes of existing entries: lost (five senses, from we got lost to lost in thought and lost time), middle (the middle seat, middle incomes), own (her own room, he cuts his own hair, with the phrases on your own and of your own), south (the south side, a south wind, with a short note on south versus southern), and top (the top shelf, a top student, top speed).

What the reviewers found and what I decided. Each entry goes to two reviewers from different companies. Their pronunciations were also checked by a three-model panel against an open pronouncing dictionary; all twenty transcriptions agreed. The reviewers raised 79 issues, 12 of them blocking (an entry cannot be published until each blocking issue is fixed or rejected with a reason). I logged 77 decisions: 59 changes made and 18 objections rejected, with none passed up to you.

The changes that mattered most:

Most rejections were requests to add idioms that by house rule belong under another word: a lost cause goes under cause, hold your own under hold, come into your own under come, and be lost for words under word. One reviewer said on your own should move to a separate pronoun entry. I rejected that under the same rule, which keeps an idiom under its keyword whatever that word's part of speech. Two other objections were aimed at the inflection notes, which a script writes from the grammar codes, so they were rejected as well.

One back-link (the matching reference a tool adds to the other entry) could not be placed by the tool: confused now lists lost as a synonym, placed by hand.

Spend. US$0.28 this cycle, for the pronunciation panel and reviewer B. Reviewer A runs on your subscription at no charge. The day's total is US$3.50 of the US$15 cap.

What did not work. The tool that stamps an entry's frequency band from the queue cannot tell homographs apart. Each new second homograph therefore got the first one's band and its "defining vocabulary" mark (the defining vocabulary is the core word list used to write definitions). Four of the five are marked as defining-vocabulary words, though the dog's bark, the tree, the side of the head and the bird's bill are not the meanings that put those spellings on the list. This is part of a known gap (wiki/notes/homograph-queue-and-inflection.md). The wiki is twenty words under its 30,000-word cap, so I did not add to that note; the problem is in the log and in NEXT.md instead.

What is next. Build cycles continue with family-box and cross-reference words in bands 1 and 2: tonight (adverb), bruise, curl, figure (verbs), and bark the verb, which now pairs with the new noun. Review cycles continue the coverage pass over the most common entries.

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

Cycle 2: review

What the cycle did. The selector chose a review cycle: a second, careful reading of entries that are already published, adding what is missing and fixing what is wrong. It read six entries.

Three come from the coverage pass, your ruling that the most common entries be reread for completeness, starting with those that have the most senses: away (adverb), power (noun), and quarter (noun). That makes fifty-five entries read in the pass so far. The other three were follow-ups the last run left in NEXT.md: breath, look, and go.

What was added:

What the reviewers found and what I decided. The two reviewers raised 98 issues, 15 of them blocking. I logged 98 decisions: 82 changes and 16 rejections, with nothing passed up to you.

One mistake was mine. For go I added two new senses, disappear and the casual say (then he goes, "No way!"). I had not noticed that the entry already covers both as subsenses: smaller numbered parts of a sense, here 2a and 10a. Both reviewers caught it, and I removed the duplicates. The lesson is now in NEXT.md: check a new sense against the entry's subsenses before adding it.

The other changes were smaller:

Rejections:

Spend. US$0.43 this cycle. The day's total is US$3.93 of the US$15 cap.

What did not work. Only the duplicated go senses, described above.

What is next. The coverage pass continues with close (adjective), write, way, and up.

What needs you. Nothing new.

Cycle 3: build

What the cycle did. The selector chose another build cycle. The next rows in the queue were common verbs that other entries already point to: entries with links to them that led nowhere until now. I took eight of them, and all eight are now published.

What the reviewers found and what I decided. All sixteen pronunciations passed the panel. The reviewers raised 77 issues, 7 of them blocking. I logged 77 decisions: 61 changes and 16 rejections, with nothing passed up to you.

The blocking issues:

Smaller fixes:

Rejections:

Four matching back-links could not be placed by the tool. I placed them by hand in approve, base, continue, and celebrate.

Spend. US$0.29 this cycle. The day's total is US$4.22 of the US$15 cap.

What did not work. Nothing new. One of my own commands got stuck waiting, because a check that looked for the reviewing program also matched its own command line; I stopped it. That cost only time.

What is next. More common verbs from the queue (mind, miss, offer, point), and the coverage pass in review cycles.

What needs you. Nothing new.

Cycle 4: review

What the cycle did. The last cycle of the run was another review cycle. It read five entries in the coverage pass: close (adjective), write, way, up (adverb), and tough. That brings the pass to sixty entries.

What changed:

What the reviewers found and what I decided. The reviewers raised 67 issues, 2 of them blocking: the write definition and the tough sense described above. I logged 53 decisions, with nothing passed up to you. Most of the minor issues were notes that repeated each other, and these were cut.

Rejections:

When I removed the rare verb author as a synonym of write, I also removed the matching link in the author entry, as house rule requires.

What did not work. Reviewer B once returned a reply with no verdicts, for up. I re-ran that reviewer, but I had edited the entry while the re-run was working, and the re-run saved over my edit. I noticed, applied the edit again, and checked it. The lesson is in the log: wait until a re-run has finished before editing the entry.

Spend. US$0.46 this cycle. The day's total is US$4.68 of the US$15 cap.

What is next. The coverage pass continues with thick, sweet, support, and stop. When it reaches about a hundred entries, I will tell you, as your ruling E5 asks. The next lint pass (the regular whole-dictionary check) is due in about seven runs.

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

All journal entries