2026-10-07 (5): closure — education, instruction, pin, section, and eight verbs cash to code
Cycle 1 of the 04:44 UTC run. Run id 20261007T044419Z-hpvo8n.
What this cycle did
The scheduler (tools/next_mode.py, the script that picks each cycle's job) chose closure: drafting entries for words the dictionary already uses somewhere (in a definition, or in a word-family box) but has no entry for yet. Closure keeps the dictionary self-contained: a reader who meets a word in a definition can look it up. The scheduler gave it this turn because closure had the largest "debt" (time since it last ran, weighted by its share of runs).
The first source, words used in definitions, offered four rows: education, instructions, pins, and sections. Three of these were plurals that the scanning tool had queued as if they were separate words. As earlier runs have done, I marked the plural rows as duplicates and drafted the real words, instruction, pin, and section, instead. The second source, words named in word-family boxes, gave eight verbs that pair with nouns already in the dictionary: cash, center, chain, chair, circle, cloud, coat, and code. Twelve entries in all, the per-cycle limit.
The nouns are the fuller entries: education has four senses (teaching and learning, the knowledge a person gains, the university subject, and the informal quite an education), a learner error about the plural, and a warning that in some languages the look-alike word also means good manners. Pin has five senses, from the sewing pin to the bowling pin, and three idioms (pins and needles, on pins and needles, you could hear a pin drop). Section has a note on choosing between part, section, and piece. The verbs are smaller, as the words are: cash and chair have one sense each, code and circle three.
What the reviewers found and what I decided
Every entry is read by two outside reviewer models, field by field; I then judge each problem they raise. This time reviewer-b (the Google model) raised nothing on any of the twelve entries, and its replies were complete, with a verdict for every field. Reviewer-a (the OpenAI model) raised 21 issues. I agreed with 15 and changed the entries; I disagreed with 6 and kept them, with the reason logged.
The useful catches: two learner errors I had written were not errors at all. The birds circled around over the field is ordinary English, and so is the third section in the book; both were removed. It also caught a "formal" label on instruction in the sense of teaching (driving instruction is everyday English), a note about cash out that was too narrow, and a translation hint on chain that could mislead a translator.
What I rejected: it wanted on pins and needles to lose its American label (British speakers usually say on tenterhooks); it wanted cross section hyphenated (the open spelling is standard American); it said Make sure the text is centered shows an adjective rather than the verb (it is the passive of the verb); and it wanted chair a meeting labeled formal (it is ordinary business English; I reworded the usage note instead, which had called the verb "fairly formal").
One pronunciation was disputed: my British transcription of education. The checking panel of models did not accept the vowel I wrote after the j sound; I changed it to the short vowel common in British dictionaries' style of transcription, and the panel then agreed (two of three).
Three definitions used words outside the defining vocabulary (the 2,394 plain words definitions are meant to use): command, badge, and orchestra. All three were rewritten in plain words.
Spend
US$0.78 for this cycle (panel review and pronunciation checks). The day stands at US$3.59 of the US$15 cap.
What did not work
Nothing failed. The scanning tool's habit of queuing plurals as words continues; it costs only a few minutes each time and is already noted for a later fix.
What is next
The run continues with its next cycle if the clock allows. The closure queue now also holds the missing family words these entries name (cashier, central, chairperson, circular, cloudy, coating, coder, decode, and others).
For the owner
Nothing new needs you from this cycle.