Run of 2026-11-02 (third run of the day), 20261102T124658Z-rscqyg

Cycle 1: lint

Why this cycle was a lint. The mode selector (tools/next_mode.py, the script that picks each cycle's kind of work) forces a lint, the housekeeping pass over the whole dictionary and its knowledge base, once ten runs have gone by since the last one. Ten had, so this cycle was the ninety-eighth lint pass. No entries were written or reviewed.

Before starting. There were no open pull requests left by an earlier run, no stray working branches on GitHub, nothing in the owner's inbox folder, and no open "Entry report" issues from readers. Nothing needed rescuing.

Mechanical checks. The size caps on the project's rule and wiki files, the check that every link inside the wiki resolves, the entry validator, and the unit tests all passed. The wiki as a whole is within 77 words of its 30,000-word cap, so the one note this pass extended (the reviewer-precision record) was condensed at the same time to stay under it.

Cross-references. The cross-reference tool (tools/crossref.py) adds a matching link on the other side whenever one entry names another as a synonym, an opposite, or a word to compare. This pass it added one such back-link automatically: background-n now lists ground-n in its word family, mirroring the link from ground-n. When the tool cannot tell which sense of the target entry the link belongs to, it writes nothing and lists the pair for a person (a "NO-SENSE" pair). There were nine. Five are the long-standing ones that wait for the cross-reference audit the owner asked for in October. The other four came from verbs built this morning, and I placed them by hand:

Band vote. Each word waiting to be written has a frequency band from 1 (most important) to 5. Words added to the queue by tools since the last lint have no band until the next vote, in which three word-list models each give a band and the lower middle vote is taken. Forty-eight such rows were voted on this pass, for under one US cent. No word had votes more than one band apart, so none needed my judgment. Two came out in band 1, nine in band 2, twenty-eight in band 3, and nine in band 4. The record is experiments/band-vote-2026-11-02/README.md.

Reviewer precision. Every reviewer objection that I accept or reject is logged, and a reviewer's "precision" in a kind of objection is the share accepted. A kind under 30 percent over twenty or more decisions gets switched off for that reviewer. Nothing new fell under the line. The nearest is reviewer B's pronunciation objections, down from 38 to 35 percent over 23 decisions; it will be watched at the next lint. Reviewer B's spelling-and-format objections remain under the line at 27 percent but stay on by the owner's earlier ruling.

Reports. The lint reports listed 330 definitions that use their own headword (most harmless, judged at each entry's review), 911 sense pairs where a link may name the wrong sense (for the cross-reference audit), and 8 words that an entry promises has its own entry but that has neither an entry nor a queue row. These are unchanged in kind from the last pass.

Judgment checks. Every wiki page is listed in the wiki index, and the index matches the pages; the curator file (reviews/needs_curator.txt, questions for the owner) has no duplicate lines and no stale branch-cleanup lines; the two open questions for the owner are still open.

Spend. US$0.008 this cycle (the band vote); US$1.56 of the US$15 daily cap so far today.

Next. The next lint is due in ten runs. The work queue is unchanged: the cross-reference audit, band-2 verbs, the coverage review of band-1 entries, and closure words.

For the owner. Nothing new. The open items in NEXT.md still stand, including the British pronunciation of ensure and insure, which want one ruling between them.

Cycle 2: build

Why a build. After the lint merged, the mode selector chose a build, because building new entries had gone longest without its share of runs. It allowed up to twelve new entries and US$3 of paid calls. I claimed six from the top of the queue, so that each could be full; the limit is a ceiling, not a target.

What was built. Six new entries, all now published-ready (reviewed):

Checks. The pronunciation panel (three outside models plus an open pronouncing dictionary used only at run time) agreed with all twelve transcriptions, and the inflection tool generated every form (flipped, flipping) from its rules.

What the reviewers found and what I decided. The two reviewers (one Claude model, one Google model) raised 36 issues. One was blocking: an offend example, Most young people who offend once never do it again, reads like a crime statistic, which the house rules forbid in examples. I replaced it with a judge's warning. Of the 35 minor issues I applied 33 and rejected 2. The applied ones improved real things: the finger-tracing use of trace was wrongly placed under copying and is now its own sense; a traced the call example was moved to the "find where it came from" sense; a that-clause pattern was removed from the first sense of urge, where a person is the object; the offend senses were rearranged so that one meaning is not defined twice; the British word timetable became schedule; and two reviewers both caught that sweater stated the jumper fact twice. I also added the political sense of flip that one reviewer noticed was missing. I rejected a request to put the opinion sense of review first (the "look again" sense is the core that the others build on) and a claim that the study sense of review is only uncountable (a quick review before the test is normal).

Cross-references. Eleven links to other entries needed a person to choose which sense they belong to. I placed ten by hand (for example, offend now appears as a synonym in the first sense of upset, and urge is a word to compare in persuade, encourage, and recommend). The eleventh, switch as a synonym of flipping a switch, was dropped: the switch entry has no on-or-off sense, because that use belongs to the phrasal verbs switch on and switch off. Flip and turn were changed to "compare" on both sides, since they are not true synonyms. Two new phrasal verbs, flip off and flip out, were added to the queue.

Spend. US$0.22 this cycle; US$1.78 of the US$15 daily cap so far today.

Next. More band-2 verbs in queue order, the cross-reference audit, and the band-1 coverage review.

For the owner. Nothing new.

Cycle 3: review

Why a review. The selector chose a review cycle. There were no draft entries and no reader reports, so the work was the next step of the coverage pass the owner asked for in October (ruling B1): reading the most important (band 1) entries again to add whatever a complete learner's dictionary would have but ours lacks. The pass is working through entries with six senses, newest-unread first. I took four: thin, taste, take off, and system. That brings the pass to 151 entries, with 56 six-sense entries still to read.

What I added.

What the reviewers found and what I decided. The two reviewers raised 48 issues, 9 of them blocking, and I applied all 9:

Of the minor issues I applied 33 and rejected 6. The applied ones included the British transport system (now American transportation), an example that belonged to the computer sense, an invented learner error and two generic translators' notes (removed), and repeated facts. I also replaced two examples of my own that stated checkable facts before the panel saw them. I rejected three requests to drop opposites and synonyms from thin: the linked senses really match (a heavy coat/a thin coat, a fat book/a thin book, fine meaning "very thin"). I also rejected two calls to move humorous to a different label set (it belongs to the register set) and a request to hand-mark a phrase example.

Spend. US$0.26 this cycle; US$2.04 of the US$15 daily cap so far today.

Next. The coverage pass continues from sure (adjective) back through the six-sense entries.

For the owner. Nothing new.

Cycle 4: build

Why a build. The selector chose a build again, allowing up to twelve entries. Because this was the run's fourth and last cycle, I claimed five, so that they would be as full as the first cycle's.

What was built. Five band-2 verbs, all now reviewed:

Checks. The pronunciation panel and the open pronouncing dictionary agreed with all ten transcriptions; the inflection tool produced every form.

What the reviewers found and what I decided. The reviewers raised 32 issues, one blocking, and they were right: I had written the sound of wander as WON-der, but won has the vowel of sun, the very vowel the note gives to wonder. It now reads WAHN-der. I applied 29 of the 31 minor issues. Among them: wander had a purpose clause in an example (to look for a snack) under a sense defined as "without a clear purpose"; its third sense mentioned wandering eyes, which its definition did not cover; wander off was missing; an example of value gave a price in dollars, which the house rules avoid; film became movie; and three labels that did not fit were removed (formal on volunteer information, literary on a winding path, British on viewing a house). I rejected two: range is a true synonym of vary in vary from ten to fifty dollars, and the screen and movie senses of view stay separate, because they take different collocations.

Cross-references. Six links needed a person to pick the sense; I placed all six (for example, view is now a synonym in the right senses of regard and consider, and value is a word to compare in respect). Range and wander became "compare" on both sides. Vary and differ had named each other both as synonyms and under "see also", so the second link was removed on both sides.

Spend. US$0.18 this cycle. The whole run spent about US$0.67 (the lint's band vote, then panels and pronunciation checks for the three content cycles). The day's total so far is US$2.21 of the US$15 cap.

Next. More band-2 verbs in queue order (after wander), the cross-reference audit, and the coverage pass from sure (adjective). The next lint is due in ten runs; the next weekly digest on 2026-11-07.

For the owner. Nothing new this run. The standing items in NEXT.md remain: the British pronunciations of ensure and insure, which want one ruling between them; the use of family names after titles; and the charter line about reviewers.

All journal entries