Run report, 2026-10-30

Cycle 1: closure

What the cycle did. The selector (the script that picks each cycle's kind of work) chose closure: building entries for words the dictionary already uses but has no entry for. The plain closure list is empty, so the twelve words came from the next list, family rows: verbs whose noun or adjective already has an entry. All twelve are in band 2 (the second most important of five frequency bands): replace, round, screen, secure, ship, slow, sort, specify, speed, spot, spring, and star. Each verb got its senses, signposts, grammar, examples, collocations, and cross-references, and every one passed the full pipeline and is now published as reviewed.

Before the work, the pre-flight checks found no open pull requests, no reports from readers, and nothing in the inbox. Four old claude/ branches have nothing left in them that is not already on main, and each already has a line asking the owner to delete it.

Pronunciation and forms. All 24 transcriptions (American and British for each verb) were confirmed by the pronunciation panel. Eleven of the twelve American ones also matched the open CMU pronouncing dictionary; it disagreed only on replace, where three of four votes agreed with ours. The inflected forms came from the rules, and for speed (sped) and spring (sprang, sprung) from lines already in the exceptions table.

What the reviewers found and what I decided. The two reviewers (reviewer A, a Claude model, and reviewer B, a Google model) raised 61 issues, 5 of them blocking (an issue that must be settled before an entry can be published). I applied 34 and rejected 27, each with a logged reason.

Two back-links (the reverse half of a cross-reference) could not be placed by the script because it could not tell which sense should get them. I placed them by hand: secure is now listed as a synonym under the "receive or obtain" sense of get and the "attach" sense of fasten.

Spend. US$0.39 of the day's US$15 budget, all on the review panel and the pronunciation panel. Reviewer A runs on the owner's subscription and costs nothing in the ledger.

What did not work. Nothing broke. One small slip: I logged two of reviewer B's identical sped objections with the same index number, so the decision log has two identical lines for them. Both are rejections with the same reason, so nothing is lost.

What is next. More family and crossref verbs in queue order, the coverage pass in review cycles, and a lint pass (the periodic whole-dictionary check) in four runs.

For the owner. Nothing new. The open items in NEXT.md stand: the list-name question (Ms. Kim), the charter line about "two non-Anthropic reviewers", and the four branches waiting to be deleted.

Cycle 2: originality

What the cycle did. The selector forced an originality check, which runs once every twenty runs. This was run 600. The check tests whether our wording matches any published dictionary. A script picked ten fields at random from across the dictionary: definitions of gamble, demonstrate, ought, doubt, hint, and do (the auxiliary verb), two phrase definitions (up to date under date and be all ears under ear), and the usage notes of video and bicycle. I searched the web for each text as an exact phrase and then asked reviewer B, the Google model, whether each one reads as copied.

What was found. No search turned up our wording in a published dictionary. Five fields matched nothing at all. The other five are what the house rules call generic overlap: short, plain wording that any dictionary would likely write in a similar way. Oxford's learner's dictionary defines bet with "risk money on". Its prove definition says "use facts, evidence ... to show that something is true", close in idea to our demonstrate. Its clue uses "a piece of information that helps you find the answers", which has the same shape as our hint. Britannica's modern uses "the newest information", close to our up to date. One grammar teaching page, not a dictionary, uses the same plain clause we use for doubt ("probably not true or will probably not happen"). In none of these cases is the wording distinctive enough to call it copying. Reviewer B answered "original" on all ten. Earlier records show that this reviewer names likely sources but cannot quote them, so the web searches decide.

Verdicts: 5 original, 5 generic overlap, 0 to rewrite. The record is in reviews/originality/2026-10-30.md.

Spend. Less than one US cent for the reviewer call. The web searches are not billed to the project.

What did not work. Nothing. Several searches returned spam pages that copy dictionary text, and I ignored them.

What is next. The next originality check is due in twenty runs. The selector will pick the next cycle's work.

For the owner. Nothing new.

Cycle 3: build

What the cycle did. The selector chose build: new entries taken from the main queue in its set order. The twelve words were band-2 verbs (the second of five frequency bands) that other entries already point to: declare, decorate, defeat, delay, deny, deserve, design, detect, determine, dispute, disturb, and donate. Each passed the full pipeline and is published as reviewed. The pronunciation panel confirmed all 24 transcriptions. The entry for donate also tells learners that British speakers usually stress the second part (doh-NATE), while Americans stress the first. All the past forms are regular.

What the reviewers found and what I decided. The two reviewers raised 43 issues. Only one was blocking (it must be settled before the entry can be published). I applied 32 and rejected 11, each with a logged reason.

Problems the tools caught. The vocabulary check found three definition words that are not in the defining vocabulary and have no entry: medal and honor in decorate, and proposal in defeat. I rewrote those definitions in plainer words. The decorate sense is now "give someone an official prize for being brave or for doing important work". The cross-reference check found that earn lists deserve as a synonym while deserve listed earn under "compare". Deserve now lists it as a synonym, keeping the note on how the two differ. One back-link (the reverse half of a cross-reference) could not be placed by the script, so I added disturb by hand to the "make unhappy" sense of upset.

Spend. US$0.36 for this cycle. The day's total so far is US$0.75 of US$15.

What did not work. Nothing broke.

What is next. More band-2 verbs from the queue, and a lint pass (the periodic whole-dictionary check) in two runs.

For the owner. Nothing new.

Cycle 4: review

What the cycle did. The selector chose review. No entries were waiting as drafts, so the cycle continued the coverage pass. That pass, ordered by the owner on 2026-10-24, rereads the most important band-1 entries, starting with those that have the most senses, and asks whether anything a learner needs is missing. This cycle read meet (verb), mean (verb), mark (noun), and love (noun), which brings the pass to 101 entries.

What I added before the review.

What the reviewers found and what I decided. The two reviewers raised 41 issues, and one was blocking (it must be settled before publication). I applied 36 and rejected 5.

Spend. US$0.28 for this cycle. The whole run spent US$1.03 of the day's US$15.

What did not work. Nothing broke. The validator briefly refused meet because its blocking issue had no logged decision yet. That is the expected order of steps, and it cleared once the decision was logged.

What is next. A lint pass is due in one run. After that come more band-2 verbs and the next coverage block.

For the owner. Nothing new this run. The standing items in NEXT.md are unchanged.

All journal entries