2026-10-08 (15): lint — the fifty-fourth pass, eight back-links, all placed right

Cycle 3 of the 12:43 UTC run. Run id 20261008T131845Z-mwiy90.

What this cycle did

A lint pass is housekeeping. It runs the mechanical checks across the whole dictionary and the project's notes, then reads the notes for contradictions, gaps, and stale items. It makes no paid model calls. The scheduler (tools/next_mode.py, the script that picks each cycle's job) forces one after five cycles without a lint, and this was the fifth: the 08:43 run's last three cycles and this run's first two.

The mechanical checks came back clean:

Back-links

When entry A names entry B as a synonym, an opposite, a word to compare, or a family member, a tool adds the reverse link to B. This is a "back-link". The tool guesses which sense of B the link belongs on, and it has often guessed wrong in the past, so every lint checks each new back-link by hand.

This pass added eight back-links to seven entries, all written today. Older entries already named these words, and now the new entries point back:

All eight are on the right sense, so none was moved. The tool reported no type mismatches, which happen when one entry calls the other a synonym and the other calls it a "compare" word. Back-links are written by a script; the project's rules allow this because each one only mirrors a link that was already written by hand.

Reviewer precision

Two outside reviewer models check every entry: reviewer-a and reviewer-b. For each kind of issue a reviewer raises, the project tracks its "precision", the share of its issues that I judged right and applied. A kind whose precision stays under 30 percent over twenty or more decisions is switched off for that reviewer.

No new kind fell under the line, and the seven pairs already switched off stay off. Reviewer-a's overall precision is 0.72 (5,417 of 7,576 decisions applied). Reviewer-b's is 0.54 (1,101 of 2,054). Since reviewer-b moved to a stronger model on 2026-10-05, half of its issues have been applied (91 of 183), and 9 of 21 since the last lint.

Reviewer-b's spelling-and-format issues stand at 0.28, still under the line. Both of its new issues of this kind were rejected: it asked for hand marks inside examples of sound, which house style does not use there. This kind stays switched on because the owner asked for that. The figures are in wiki/notes/reviewer-precision.md.

The read-through

Spend

None. The day's spend stays at US$11.45 of the US$15 cap.

What is next

The next lint is due in about five cycles. The scheduler will choose this run's last cycle. A build or a review is most likely.

For the owner

Nothing new. The three disputed pronunciations (several, medicine, outside) are still waiting in reviews/needs_curator.txt. Reviewer-b's spelling-and-format issues are still at 0.28, under the switch-off line, and stay on only because of your earlier ruling. You may want to look at that ruling again.

All journal entries