Run report, 2026-10-27 (scheduled run 20261027T064413Z-tm5zwm)

Cycle 1: build

What the cycle did. The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work) chose build: writing new entries from the queue of words still to be written, with a ceiling of twelve entries and a paid-call budget of US$3. The reason it gave was the usual one, that build had the highest "debt" among the eligible modes; closure (writing entries for words the dictionary already uses in its own definitions) was still blocked because milestone 1 had six rows left.

Milestone 1 is the set of about 2,400 words the dictionary uses to write its definitions; the owner ruled that it must be finished before anything else is built. The twelve words claimed this cycle were the last of it plus six pronouns that point back to existing determiner entries: the prefixes out-, over-, trans-; the suffix -dom; the combining forms micro- and tele-; and the pronouns half, many, more, most, much, and no one. Milestone 1 now has no queue rows left. Five homographs (a second, unrelated word with the same spelling, such as bark of a tree versus a dog's bark) are still to be claimed by hand, as NEXT.md lists; the selector now points to closure.

Pre-flight found no stranded pull requests, no reader reports, nothing in the owner's inbox, and three old work branches with nothing unmerged on them, already listed for deletion in reviews/needs_curator.txt.

Each entry went through the full pipeline: the schema check, the inflection tool, the pronunciation check (three outside models vote on each IPA transcription, and the open CMU pronouncing dictionary is consulted for American English where it has the word), and the two-reviewer panel. Reviewer A is a Claude model; reviewer B is a Google model. Both read every field and give a closed verdict on each.

What the reviewers found and what I decided. The panel raised 71 issues, 10 of them blocking. I applied 53 and rejected 18, each with a logged reason in reviews/decisions.jsonl; nothing needed escalation. The main changes:

The rejections were mostly requests for idioms that already live in a sister entry (more or less in the determiner more, what's more and so much for in the adverbs) or that belong under their first verb (make the most of under make), plus seven requests to bold the affixed word in trans- examples, which house practice leaves unmarked.

The pronunciation panel agreed on everything except trans-, where I had written tranz and two of the three votes said trans. I changed the transcription to trans, kept tranz in the plain-English note as the form often heard before a vowel, and it then passed.

Spend. US$0.34 this cycle; the day stands at US$1.32 of the US$15 cap. Reviewer A runs on the owner's subscription and costs nothing in the ledger.

What did not work. The definition-vocabulary check caught further in two core-idea sentences and millionth in micro-; all three were reworded in plainer words before the entries were marked reviewed.

What is next. With milestone 1 finished, closure opens for bands 1 and 2. For the owner: nothing new needs a ruling this cycle; the open items in NEXT.md stand.

Cycle 2: closure

What the cycle did. With milestone 1 finished, the selector chose closure for the first time since the owner's ruling of 2026-10-24 paused it. Closure writes entries for words the dictionary already uses in its own definitions or names in its cross-references. Under that ruling it takes only words in bands 1 and 2 (the two most important of the five frequency bands). It allowed twelve entries and US$3 of paid calls.

The closure rows (words found in existing definitions) gave a poor start: four of the six were inflected forms queued as if they were headwords (polished, candles, eyebrows, nails). This is a known weakness of the vocabulary tool, which guesses a word's base form and part of speech and sometimes fails to strip an ending. I marked the four rows as duplicates and claimed the real words instead: candle, eyebrow, and nail. Polish is in band 3 and has to wait. With the closure rows used up, I took five family rows (words named in an existing entry's word-family list): the verbs please and step, and the nouns base, board, and boyfriend. Diamond and lemon complete the set. Midway through, the vocabulary gate refused baseball in two definitions (in diamond and base), because no entry existed for it. I wrote that entry too, for eleven in all. The note on the vocabulary tool's weakness now records this third recurrence, and says a fix is worth a logged decision in a later run.

What the reviewers found and what I decided. The panel raised 49 issues, 3 of them blocking. I applied 42 and rejected 7.

Rejected: a request to call whatever she pleases transitive (the verb stands for an omitted to do), a request to file burn the candle at both ends under burn (the house rule takes the first noun), a request to reorder the senses of board, and two reviewer objections to my nail adaptation note. That note follows the style guide's own model for bank.

The pronunciation panel voted three to one for a three-syllable American diamond (DY-uh-mund). That is now the main form, and the two-syllable form is a variant.

What did not work. The cross-reference tool placed a back-link from walk to step on the wrong sense of walk (its "explain step by step" sense, matched by the shared word step). I moved it by hand. It also could not choose a sense of basis for the link from base, so I placed that one by hand too. The vocabulary gate then refused bat in the new baseball definition. Bat has no entry yet, so the definition now says a long stick.

Spend. US$0.36 this cycle; the day stands at US$1.67 of US$15.

What is next. Closure continues from family and crossref rows in bands 1 and 2. Lint is due in three runs.

Cycle 3: closure

What the cycle did. The selector chose closure again, with the same limits: twelve entries and US$3 of paid calls. No closure rows (words found in existing definitions) are left in bands 1 and 2, so the claim tool took the next ten family rows, words named in an older entry's word-family list. All ten are band-1 nouns, the most important band: call, classroom, cover, gift, gold, homework, ice cream, meeting, mom, and mum. The heavy ones, call (six senses) and cover (nine after review), were drafted with the matching verb entries open beside them, so the noun does not repeat the verb's senses or idioms.

Each entry went through the full pipeline. The pronunciation panel verified every transcription, including the two stress patterns of ice cream: American English stresses the first word and British English the second. The pronunciation note says so.

What the reviewers found and what I decided. The panel raised 39 issues, 4 of them blocking. I applied 30 and rejected 9.

Rejected: heart of gold (it already lives under heart, the first noun), marking struck in a phrase example, a "dated" label on the narrowed visit sense, and treating mum as a spelling variant of mom. They are separate words linked to each other.

What did not work. After review, the vocabulary gate refused occasion, property, and charity in the reworded definition of gift, and occasion in meeting. An earlier check by the same tool had passed these words, and I could not see why. The gate is what continuous integration runs, so I trusted it and reworded all four. The lesson for later cycles is to run the gate itself (--gate --changed) after the last edit, not the plain report.

Spend. US$0.31 this cycle; the day stands at US$1.98 of US$15.

What is next. More family rows remain in band 1 (reading, return, run, set, and others). Lint is due in two runs.

Cycle 4: closure

What the cycle did. The selector chose closure a third time, with the same limits (twelve entries, US$3). The next family rows (words named in an older entry's word-family list) included run and set as nouns. Both are very large entries, and they deserve a cycle with fewer companions, so I left them in the queue. I claimed eight smaller band-1 words instead: the nouns reading, return, share, shopping, start, swimming, and writing, and the adjective amazing. Each was drafted with the matching verb entry beside it, so the noun holds only noun uses.

This time I ran the vocabulary gate on the drafts before review, not after. It caught poem, aloud, and fright at once, and later tennis, net, and forth in a sense added during review. All were reworded in plainer words.

What the reviewers found and what I decided. The panel raised 41 issues, 9 of them blocking. I applied 32 and rejected 9.

Rejected: the lion's share and the point of no return, which the house rule files under their first nouns, lion and point; a flap symbol in American writing (the house transcription writes a plain t, and the panel verified it); markings inside phrase examples; and reviewer A's objection to the grammar code comparative with more on amazing, which is a settled house rule.

Spend. US$0.29 this cycle; the day ends this run at US$2.27 of the US$15 cap.

What is next. The selector's lint pass (a periodic check of the whole dictionary and wiki) is due next run. After that, closure continues with run, set, and the remaining family rows.

For the owner. Nothing new needs a ruling. Milestone 1, the defining vocabulary, was finished in cycle 1 of this run. The ruling E5 report waits until the first coverage-review block of about 100 entries is done.

All journal entries