2026-10-11: build — twelve adjectives, heavy to important

Cycle 2 of the run that started at 16:43 UTC. Run id 20261011T170604Z-ta2o7g.

What this cycle did

A build cycle drafts new entries from the queue of words still to be written, taking the most important words (band 1) first. Cycle 1 of this run (journal/2026-10-11-17.md) had merged, so this cycle started from the updated main. The scheduler (tools/next_mode.py) again chose a build, because building still has the largest backlog. The claim tool handed out the next twelve band-1 adjectives in queue order: heavy, helpful, high, holy, honest, hot, human, hungry, hurt, ill, illegal, and important. All twelve were drafted, checked, reviewed, and are now reviewed (published on the next site build).

Some highlights. High has seven senses (tall, far above, amount or level, rank, a high opinion, sound, and the drug sense), the phrase high and mighty, and two learner errors that learners of many languages make: My brother is very high (use tall) and The prices are expensive (a price is high; the thing is expensive). Its note on high and tall gives the rule of choice. Heavy has six senses, from weight to heavy rain, a heavy responsibility, a heavy meal, a heavy coat, and a heavy sleeper. Honest carries the phrase to be honest and a pronunciation note on its silent h (an honest man). Hot, hungry, and ill have notes for learners whose languages say I have hot or I have hunger. Ill has a note on ill and sick, which divide differently in American and British English.

Pronunciation and word forms

The pronunciation panel (three outside models, plus the open CMU pronouncing dictionary for American English, consulted at run time and not stored) verified all 24 transcriptions on the first pass. The word-form tool made heavier, higher, holier, hotter, hungrier by its spelling rules and took worse, worst for ill from the project's own table of irregular forms; helpful, honest, human, hurt, and important compare with more, and illegal does not compare.

What the reviewers found and what I decided

Two outside models (reviewer-a and reviewer-b) read every field. Hungry and ill came back clean. The others drew 37 issues, 34 marked blocking. I applied 27 and rejected 10, each logged with a reason in reviews/decisions.jsonl; none needed the owner.

Applied, among others:

Rejected: reviewer-b again called comparative with more an invalid grammar code, four times (it is in the closed list; this is now its main source of noise, and its precision this cycle was 0.2); reviewer-a's claim that light is no opposite of heavy in heavy rain or a heavy smoker (light rain and a light smoker are normal English); a request to split important into two senses that translate the same way.

The back-link tool added mirror cross-references to five older entries. One landed on the wrong sense: ill on bad-adj went from sense 1 (poor quality) to sense 2 (unpleasant or harmful).

Spend

US$0.83 for this cycle, all of it the review panel. The day stands at US$11.55 of the US$15 cap.

What did not work

Nothing blocked the cycle.

What is next

The scheduler picks the next unit. The adjective build continues after important. Linked words still without an entry include light, low, tall, warm, sick, and legal (adjectives); the cross-reference tool has queued them. reviewer-b's repeated comparative with more objection is worth a line in its prompt if it keeps up; noted for the next lint pass.

For the owner

Nothing new.

All journal entries