2026-09-21-4
This was a build run (a "build" run drafts new dictionary entries, as opposed to a "review" run, which re-checks existing ones). I added ten new entries: the nouns accident, action, advantage, and age; the verbs agree, allow, and afford; and the adjectives angry, aware, and available. All ten are common, everyday words from the project's "band 1" list — the highest-priority vocabulary a learner needs. The dictionary now has 250 entries, all fully reviewed.
The previous run's handoff note said the two heaviest remaining band-1 words, on and over (prepositions with many senses each, like run or hand), were still being deferred as too large for a single run. I kept deferring them and instead picked ten more tractable words to widen the mix of nouns, verbs, and adjectives already drafted.
Each entry went through the project's full pipeline: I drafted the definitions, examples, grammar notes, and so on myself; an automated tool generated the plural/verb-tense forms; another tool checked the pronunciation transcriptions I wrote against a panel of AI models and a pronunciation dictionary; and finally two independent AI "reviewer" models checked every field of every entry against a checklist. I then read every issue they raised myself and decided whether to fix it, reject it, or flag it as an open question — I never apply or dismiss a reviewer's comment without reading it.
The reviewers raised 31 issues, 17 of them serious enough to block publication, and I applied fixes for 30 of the 31 (one reviewer was right almost every time this run, the other about seven times out of eight — both good scores). The most substantial catch: my entry for age was missing a very common informal meaning — "ages," as in "I haven't seen you for ages," meaning a long time — which I added as a third sense. A reviewer also caught that my definition of the phrase "of age" was too narrow (I had defined it as being old enough for one particular thing, like voting, when it actually means having reached legal adulthood generally); I corrected that too.
Two more things worth flagging: first, a reviewer said my entry for agree used an example ("the two sides agreed a new contract") that's natural in British English but unusual in American English, without a label to warn a learner. Rather than just slap a regional label on it, I swapped in the version that works in both dialects ("agreed on a new contract") and explained the difference in prose instead — a cleaner fix. Second, one of the two reviewer AI models failed to respond validly on one entry (a known, already-documented glitch); I simply asked it to try again, and it worked cleanly the second time.
I rejected two reviewer suggestions: one claimed a certain grammar note wasn't in the project's approved list of patterns, but it was — I checked the source list directly. The other was a nitpick on an optional field that didn't actually need fixing.
While drafting, I noticed three of my sentences used words like "history" and "permission" that don't yet have their own dictionary entries and aren't on the core word list — the project's style rules don't allow that in certain key fields, so I reworded those sentences with simpler words instead of leaving a rule violation. Fourteen other words used more loosely in examples (like "vacation" or "reckless") got queued for someone to write entries for later, which is normal and expected.
Total spend today: about $2.31 of a $5 daily budget; this run itself cost about 29 cents.
Nothing needs the owner's attention beyond what's already listed as open in NEXT.md: enabling GitHub Pages for the site, and two small style questions about word splitting and formatting marks that have working assumptions in place already.