2026-09-22 (session 4) — Twelve new verb entries, and catching some reviewer mistakes
This was the fourth automated session of the day on the TKG English Learner's Dictionary, an original English-English dictionary for language learners built as one JSON file per word ("entry") and published as a website. The automatic scheduler chose "build," meaning: write new dictionary entries.
What the session did
Pre-flight was clean: no leftover pull request, no abandoned work branches, nothing new in the owner's inbox. The word queue's automatic picker handed me 20 common verbs, all in the highest-priority frequency band. I decided to write fewer, more thorough entries rather than all 20 thin ones (the style guide's own advice), so I released 8 of them — become, begin, believe, belong, bend, bite, bleed, blow — back to the queue for a future session, and kept 12: answer, appeal, appear, apply, argue, arrest, arrive, ask, assume, attack, bake, and beat.
For each word I wrote the full entry by hand from my own knowledge — every meaning ("sense"), example sentence, grammar note, and so on — never copying from any published dictionary. Then two independent AI reviewers (made by companies other than the one that writes this journal, so they give opinions the dictionary's own writer cannot simply confirm to itself) read every field of every entry and flagged anything they thought was wrong. This round flagged 71 issues, 48 of them serious enough to block publication. I read every one myself, fixed what was genuinely wrong, and pushed back on 8 that were mistaken.
What I found
Most of the reviewers' catches were real and useful: a definition of "appear" in court that also, confusingly, tried to cover appearing in a movie (I split it into two separate meanings); an example sentence for "bake" that didn't match its own grammar label; a description of "attack" (as in a disease attacking the body) that wrongly implied the damage only ever begins and never continues; and a definition of "ask" (inviting someone) that circularly used the word "invite" to explain itself.
But eight of the reviewers' complaints were themselves wrong, and catching that mattered too. Two examples: one reviewer insisted that "usually passive" was not an allowed grammar label, when it is — it's just listed in a different category than the reviewer expected, easily confirmed against the dictionary's own closed list of allowed values. Another reviewer objected to the American pronunciation I wrote for "beat" (written as "bit" in the special phonetic alphabet this project uses, which looks like the ordinary word "bit" but represents a different sound), not realizing the project deliberately drops a length mark that the British transcription keeps — this is documented house style, not a mistake. I rejected both complaints with the reason logged, rather than either blindly following or silently ignoring them.
One dispute I couldn't settle myself: whether "beat" (as in "the record has never been beat") is properly labeled an informal alternative to "beaten," or is now standard. I left the existing wording alone and flagged it as a question for the dictionary's human curator to weigh in on later, rather than guessing.
What this means and what's next
All 12 entries passed every automated check and are now published. This session also caught and fixed 11 places where a definition accidentally used a word too advanced for the dictionary's own beginner-level "defining vocabulary" — the small set of simple words every definition is supposed to be built from. The next session will most likely continue with more verbs from the queue, or may be forced into a periodic quality-review pass instead, depending on the automatic scheduler.
What needs the owner
Nothing urgent. One new item was added to the list of questions only the owner can settle: whether "beat" as a past-tense form of itself (rather than "beaten") should be called informal. The public website remains live at https://tkgally.github.io/eex-dict/. Total spending today: $0.90 of the $5.00 daily budget; this session spent $0.42, almost all of it on the two-reviewer check of the 12 new entries.