2026-10-18, cycle 1 of the 12:43 UTC run: the thirty-ninth originality check

What this run did

This was an originality cycle. Every tenth run, the scheduler (tools/next_mode.py) forces a check that our wording has not drifted toward a published dictionary's. This was run 460, so the check was due. The tool (tools/originality_check.py) picked ten fields at random from finished entries: eight definitions, one phrase definition, and one usage note. For each, I searched the web for the exact wording and looked at whether any published dictionary or word list shows the same or nearly the same text. I also asked one outside reviewer model (reviewer-a) whether each field reads as copied. The verdicts are in reviews/originality/2026-10-18.md.

Before the check, the usual start-of-run sweep found no open pull requests and no leftover work on old branches; the four old claude/* branches are already listed for pruning in reviews/needs_curator.txt. The inbox was empty.

What the searches found

Three fields are original: no published wording was close. These are copy in the sense one of many books ... that are exactly the same, the usage note on on the beach and at the beach, and on the preposition for buses, trains, planes, and horses.

Two are generic overlaps: short, plain definitions that any dictionary would write in much the same way. Shoulder as meat from the top of an animal's front leg is the only natural way to name that cut. The party idiom (a person who loves going to parties and enjoys them very much) shares the plain idea "loves parties" with other dictionaries but not their wording.

Five needed a rewrite, the most in any check so far:

All five were rewritten by hand, field by field. No script touched them.

What the reviewers found and what I decided

A rewrite counts as a review of the entry, so the four changed entries went to both reviewers again (the "panel"). They raised 7 issues, 4 of them marked blocking. I applied 1 and rejected 6.

Applied: reviewer-a said the advice at the end of the culture note on lift (Choose the word for the readers) was vague. It now says Use the word your readers or listeners will expect.

Rejected:

What reviewer-a said about copying

reviewer-a answered copied for eight of the ten fields, each time naming the same learner's dictionary and "quoting" our own wording back, once with bicycle for our bike. Three of those quotes did match real wording, but in other dictionaries than the one it named; the rest were not found. This continues the pattern recorded in wiki/notes/reviewer-noise.md (updated with this check): its quotes echo the input, and the web search alone decides the verdicts.

Checks

All entries pass validation (2,530 entries, 0 errors), the word caps, the link check, the unit tests, the defining-vocabulary gate, and the cross-reference gate. Every pronunciation and inflection on the four entries is verified.

Spend

About US$0.31 (the reviewer question, US$0.003, plus the panel on the four rewritten entries). The day stands at about US$4.56 of the US$15 cap.

What is next

The next originality check is due at run 470. The scheduler chooses the next cycle; a lint pass is due in a few cycles.

What needs the owner

Nothing new. Five rewrites in one sample of ten is higher than usual (the last checks found one or two). Most matches were with very short, plain definitions, but two were word-for-word. If later checks keep finding this rate, a wider sweep of older definitions may be worth a decision.

All journal entries