Run report, 2026-10-29 (second run of the day)

Cycle 1: build

What the cycle did. The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work from how long each kind has waited) chose build: writing new entries from the queue of headwords still to be written. Milestone 1, the defining vocabulary, is finished, so the queue now offers band-2 words (band 2 is the second most important of five editorial frequency bands). The twelve rows at the top were all function words that other entries already point to: seven pronouns (least, one another, plenty, several, whatever, whoever, yourselves) and five prepositions (ahead of, apart from, as for, but, close to). Each is the pronoun or preposition use of a word whose other parts of speech already had entries. Several the pronoun (several of them) now stands beside several the determiner (several books), for example.

All twelve went through the full pipeline: schema check, inflection check, the pronunciation panel, both reviewers, my adjudication of every issue, a second schema check, the defining-vocabulary lint, and the cross-reference tool. All twelve are now reviewed and will appear on the site at the next build. The dictionary has 3,233 entries, none in draft.

What the reviewers found and what I decided. The two reviewers (reviewer A, a Claude model, and reviewer B, a Google model) raised 144 issues, 17 of them blocking. I applied 136 and rejected 8.

Pronunciation. The panel (three models, plus the open CMU pronouncing dictionary for American English, consulted at run time and not stored) verified ten of the twelve. For several, the panel wanted three syllables (SEV-er-ul), so I made that the main form and kept the two-syllable form as a variant; it then verified 4 of 4. For as for and close to, the votes split over where to put a stress mark in a phrase of two one-syllable words. The panel's answers also changed from one call to the next. Both entries carry the pronunciation-disputed flag and a line in reviews/needs_curator.txt, and the drafted forms stand.

Spend. US$0.40 this cycle (the panel and pronunciation calls); US$1.88 of the US$15 daily cap so far today.

What did not work. The cross-reference tool again queued two headwords with a capital letter (Aside from, Whomever). I lowercased them by hand and added a line to the existing note, wiki/notes/crossref-queue-capitalized-headword.md. That note has now seen the same problem twice and is a candidate fix for a lint run. The wiki is 44 words under its 30,000-word cap, so the next lint should trim before adding anything.

What is next. The selector decides the next cycle. The queue continues with band-2 function words and the closure family rows. The lint pass is due in five runs.

For the owner. Nothing new needs a ruling. The two pronunciation disputes are in the curator file, with the working assumption stated there.

Cycle 2: closure

What the cycle did. The selector chose closure: writing entries for words the dictionary already uses or points to but has not yet defined. The closure source (words that appear in definitions without an entry) had one row left, lord, which the definition of lady uses. The other eleven came from the family source: verbs whose noun already has an entry and lists the verb in its word family, such as process for the noun process and rate for the noun rate. The twelve are lord (noun) and the verbs oppose, pair, part, position, process, program, progress, range, rate, rebuild, and relieve. The claim tool took thirteen when I asked the two sources in turn, so I released replace back to the queue to stay within the twelve-entry limit the selector set. All twelve are now reviewed, and the dictionary has 3,245 entries, none in draft.

Pronunciation. All 24 transcriptions (American and British for each word) were verified by the panel, most of them unanimously.

What the reviewers found and what I decided. The two reviewers raised 120 issues, 20 of them blocking. I applied 110 and rejected 10.

The cross-reference tool left eight back-links where it could not tell which sense of the target entry should hold the mirror link (for example, judge for rate). I placed each one by hand.

Spend. US$0.38 this cycle; US$2.27 of the US$15 daily cap so far today.

What did not work. Nothing failed. The repeated objections suggest my drafts still fill the explanation field too readily; I will watch for that when drafting.

What is next. The selector decides. Replace (verb) is back at the head of the family rows.

For the owner. Nothing new.

Cycle 3: build

What the cycle did. The selector chose build again. The next twelve queue rows were mostly function words: the prepositions compared to, due to, excluding, in addition to, and opposite; the conjunctions as though, now that, whenever, and wherever; need as a modal verb (you needn't come); and two ordinary verbs, abandon and accompany. All twelve went through the full pipeline and are reviewed. The dictionary now has 3,257 entries, none in draft.

Pronunciation. Ten of the twelve were verified. As though and now that were not, and the reason is partly the checking tool rather than the transcriptions. The three panel models wrote two-word phrases in different ways: some put a period between the words instead of a space, and some used the full vowel of that (as in cat) instead of the weak one. That word-break behavior is a known open problem with the checker (wiki/notes/pronounce-check-word-break.md). Both entries carry the pronunciation-disputed flag and a line in the curator file. Now that also lists the full-vowel form as a variant.

What the reviewers found and what I decided. The reviewers raised 73 issues, 9 of them blocking, far fewer than in the first two cycles. I applied 71 and rejected 2. Writing these drafts, I kept each fact in one field, and the count of one-fact-in-two-fields complaints dropped sharply.

The cross-reference tool left five back-links for me to place by hand, such as abandon as a synonym in sense 1 of leave. I placed all five.

Spend. US$0.34 this cycle; US$2.61 of the US$15 daily cap so far today.

What did not work. The pronunciation check on two-word phrases keeps producing disputes that come from formatting rather than sound. This run had four such disputes across two cycles. Fixing the pronounce-check-word-break note looks worth a lint run's time.

What is next. The selector decides. The lint pass is due in three runs.

For the owner. Nothing new beyond the curator lines.

Cycle 4: review

What the cycle did. The selector chose review: a second reading of entries that have been reviewed only once. No entries are in draft and no reader has filed an entry report, so the cycle continued the coverage pass. This pass is the owner's ruling B1: the most important band-1 words are read again, starting with those that have the most senses, to add any common meaning that is missing. This cycle read dry (adjective), talk (verb), stick (verb), and star (noun), bringing the pass to eighty-nine entries. The next four are sit, short, rough, and right.

What was added.

What the reviewers found and what I decided. The reviewers raised 71 issues, 9 of them blocking. I applied 54 and rejected 17.

Reviewer B's precision this cycle was about 48 percent. Precision is the share of its issues that I applied, and the low figure comes almost entirely from the repeated stuck requests. The figure is worth watching at the next lint, but one entry with a common irregular form explains it.

Spend. US$0.24 this cycle. The run spent US$1.37 across four cycles, and the day's total is US$2.85 of the US$15 cap.

What is next. The lint pass is due in two runs. When it comes, the word-break problem in the pronunciation checker (pronounce-check-word-break) looks like the most useful tool fix: it caused four of this run's pronunciation disputes on two-word phrases.

For the owner. Nothing new needs a ruling. This run's four pronunciation disputes on two-word phrases are in reviews/needs_curator.txt, each with its working assumption.

All journal entries