Run report, 2026-09-25 (sixth run of the day): a second reading of twenty pronouns

What this run did

This was a scheduled, unattended run of the dictionary project. The scheduler (the script tools/next_mode.py, which picks each run's kind of work by how long each kind has waited) chose review: a second reading of entries that have been checked only once. No entries were waiting in draft status (drafted but not yet checked), so the run took the twenty oldest entries that had been through only one round of the review panel. The panel is two language models from other companies, called reviewer-a and reviewer-b, that read every field of an entry and report problems with a fixed set of verdicts. All twenty were personal pronouns and similar words written on 2026-09-19: everybody, everyone, everything, few, he, her, hers, herself, him, himself, his, I, itself, me, mine, myself, ours, ourselves, she, somebody.

Before starting, the run checked for unfinished work from earlier runs: there was no open pull request and no leftover branch, and the inbox for the owner's instructions was empty.

What the reviewers found and what I decided

Reviewer-b's reply for three entries (everybody, everything, him) came back cut off and could not be read, so I sent those three to reviewer-b again; the second replies were complete.

In all, the reviewers raised 22 issues, 15 of them marked blocking (meaning the entry should not stand without a decision). I read each against the entry and decided: 14 applied, 8 rejected, none sent to the owner.

The real problems:

Rejected: flags about pronunciation conventions the project has already settled (secondary stress on -one in everyone, a stress mark placed after a syllable break, and a claim that American fju lacks a vowel it plainly has), one flag against the phrase "and so on", and two flags that the reflexive rule was stated too absolutely, when the notes were already limited to the reflexive use.

While these entries were open I also fixed something the reviewers did not flag: four reflexive entries showed wrong forms such as come with himself in italics inside notes, which the house style forbids (wrong forms belong only in the learner-error box). I reworded them.

Something found for later

A scan showed 45 one-sense entries carrying a signpost, against the style guide. The validator does not check this. Because a signpost is written content, no script may clear it; the remaining 42 will be fixed by hand when a review opens them. I wrote this up in wiki/notes/one-sense-signposts.md; a validator warning could be added in a later run.

Spend

US$0.47 for the panel calls this run; US$2.33 today, under the US$5 daily cap.

Checks

All the closing checks passed: the entry validator, the size caps, link checking, the unit tests, the defining-vocabulary check (no definition uses a word outside the defining list without its own entry), and the cross-reference check.

What is next

The scheduler decides. Build runs continue through the band-1 verbs after gather. Review runs continue with the next pronouns (someone to us); about 322 entries remain at one review round. A lint run (a housekeeping pass over the whole dictionary) is due after one more run.

For the owner

Nothing new needs you. The three open questions (splitting be/have/do/one by part of speech, the look of the inline marks, and whether infinitive to gets an entry) are unchanged in wiki/open-questions.md.

All journal entries