2026-10-12, run at 20:44 UTC, cycle 2: the thirty-fourth originality check
What this cycle did
This was an originality cycle. Its job is to make sure no wording in the dictionary was copied from a published dictionary. The scheduler (tools/next_mode.py, the script that picks each cycle's job) forces one every tenth run, and this was run 410.
The check tool (tools/originality_check.py) picked ten fields at random from the 2,262 entries: seven definitions and three usage notes. For each one I searched the web for the exact wording and looked only at whether a published dictionary or word list used the same words. I also put a fixed question to one reviewer model, as the checklist asks. Each field then got one of three verdicts:
- original: no match found;
- generic-overlap: a short, plain wording that any dictionary would write the same way;
- rewrite: close enough to a published dictionary's own wording that it must be rewritten.
Results
Five fields were original: the honey and sun usage notes, the thumbs down gesture, in someone's good books, and the "very exact and organized" sense of military. Three were generic overlaps: the spoon definition (every dictionary describes a bowl at the end of a handle), justice as "the quality of being fair or reasonable", and the phrase meaning "unpleasant to look at". All three are short, plain wordings.
Two needed a rewrite:
- spark, sense 3, read "a small amount of a feeling or a quality". A well-known learner's dictionary has the same words in a slightly different order. It now reads "a very small sign that a feeling or quality is there". My first try used slight, which is not in the defining vocabulary (the list of about 2,400 plain words that definitions must use), so I changed it to small.
- out of town (a phrase in the entry town) read "away from the place where you live, especially for a short time". A learner's dictionary defines it as away from the town or city where you normally live, usually for a short period. Ours was not word for word the same, but it followed that frame closely. It now reads "not at home because you have gone to another city or area for a few days".
Earlier checks rewrote egg, weekend, confuse, and lung on the same grounds.
The reviewers
I sent both rewritten entries back to the two reviewer models. reviewer-b found nothing. reviewer-a raised seven issues, all marked blocking. I accepted four:
- On spark, "be the spark for" was listed as a verb + noun collocation. It is really a fixed phrase, so I moved it to a "phrase" collocation group.
- On town, the phrase in town was defined as "staying in or visiting the place". That missed people who are simply present there. It now reads "present in the town or city where you are or are talking about, often for a visit".
- On town, the usage note gave in town as an example of town meaning the town center. That clashed with the phrase in town defined just below it. I replaced the example with walk to town. This fixed two of reviewer-a's issues at once: the usage note and the clash between the two fields.
I rejected three:
- The explanation "Usually without a or the" already allows for the article.
- "Let's go and paint the town red" is natural English.
- downtown in the word family is allowed: the style guide accepts related words there, compounds included.
Each decision is logged in reviews/decisions.jsonl.
The reviewer's answers to the originality question
reviewer-a (with its hidden reasoning turned off) answered copied five times, cannot tell three times, and original twice. For spark its quotation matched what the search found. For good books and out of town it "quoted" our own text back to me, a pattern seen in earlier checks. Its other two copied answers were for short, plain phrases. The verdicts rest on the searches.
Spend
US$0.23: under a cent for the originality question and US$0.22 for the panel on the two entries. That went over this cycle's planned ceiling of US$0.10. The panel runs both reviewers over every field of the whole entry, and town is a long entry. The day's total is US$13.96 of the US$15 daily cap, which leaves about US$1.04.
What did not work
Nothing failed. The panel cost more than the selector planned; a later run may want a cheaper way to recheck a single rewritten field.
What is next
The next originality check is due at run 420. The build queue continues with band-1 adjectives after sick.
For the owner
Nothing new needs a decision. The record is in reviews/originality/2026-10-12-3.md.