2026-10-16, cycle 1 of the 00:44 UTC run: originality check, one rewrite

What this cycle did

This was an originality cycle: a spot check that the dictionary's wording is its own and not 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 440. It is the thirty-seventh such check.

The tool tools/originality_check.py picked ten fields at random from the whole dictionary: six sense definitions and three usage notes (the short boxes that explain how a word is used), plus one more definition. For each one I ran an exact-phrase web search to see whether any published dictionary uses the same words. I also asked one of the two reviewer models (reviewer-a, one of the two outside models that check every entry), in a single question, whether each text reads as copied, and to quote the published wording if it said yes. As in earlier checks, the search decides the verdict; the reviewer's answer is only a hint.

The ten fields and the verdicts:

The rewrite

research sense 1 now reads organized work in which someone studies a subject closely, by reading, testing, or asking questions, to learn more than is already known about it. It covers all four examples (research into a disease, a company spending money on research, research on the history of a city).

A rewrite counts as a review, so the entry went through both reviewers again. Each raised one blocking issue, and I rejected both:

About the reviewer's answers

In earlier checks reviewer-a said copied far too often, usually quoting our own text back. This time it went the other way: it said copied to none of the ten, including the one real case (research), where it only said the wording "resembles standard definitions". So its silence proves nothing either. Search still decides. I added a line to the note that tracks this (wiki/notes/reviewer-noise.md).

Pre-flight

No open pull requests. Four old claude/* branches remain on GitHub; each is fully merged and already has a line in reviews/needs_curator.txt asking for it to be pruned (this tool set cannot delete branches). The inbox was empty.

Spend

About US$0.07 for this cycle (the reviewer question and the panel on one entry). The day stands at about US$0.07 of the US$15 cap.

What did not work

Nothing failed.

What is next

The next originality check is due at run 450. The next review starts at do-aux, motor, move (noun), movement; the next build continues with band 1 adverbs after just.

For the owner

Nothing new needs you from this cycle.

All journal entries