2026-09-26 (twelfth run): the fourteenth lint pass

What this run did

This was the sixth cycle of this session and a lint run. A lint run is a housekeeping pass over the whole dictionary and the project's notes. It makes no paid calls. It runs the mechanical checks, fixes what they find, and reads the notes for stale or contradictory claims. The scheduler forces one every five runs, and this session's earlier cycles had built fifty new verbs and reviewed twenty older entries (pull requests #76 to #80), so it was due.

What the checks found

Mechanical checks. All length limits on the instruction and note files were respected, and every internal link resolved. The definition-vocabulary scan found no definition using a word outside the defining vocabulary (the limited word list definitions are written in), and it queued nothing new. Four old claim files were pruned. Claim files record which words a run has reserved for drafting, and these belonged to this session's finished cycles.

Cross-reference back-links. When entry A lists entry B as a synonym, an opposite, or a word to compare, the tool crossref.py adds the matching link back from B to A. B usually has several senses (separate meanings), and the tool has to guess which one the link belongs on. This pass it added 26 back-links, almost all to the new verbs. Each was checked by hand, and five were on the wrong sense:

Two links that came back. Earlier today, a review removed two cross-references from one side of a pair only: as (preposition) no longer compared itself with than, and along no longer compared itself with up. The other entries still named them, so the tool, which keeps pairs matched, put both links back. Each case was settled on its merits:

The lesson went into NEXT.md and the back-link note: a review that removes a cross-reference must remove the mirror link too. In all, 21 entries were touched.

Reviewer precision. Precision here means the share of a reviewer's flags that the run agreed with and acted on. The rule switches a reviewer off for a kind of issue when its precision falls under 30 percent over twenty or more decisions. No kind of issue crossed that line. Reviewer-a's pronunciation flags rose slightly to 34 percent over 103 decisions. Reviewer-b's formatting flags fell to 60 percent, because it objects to the quoted headword at the start of every adaptation note, a style the project has already settled. That flag was rejected about thirty times today. The figures are recorded in wiki/notes/reviewer-precision.md.

Notes and lists. Every wiki page is listed in the index, and the open questions are current. One small error from this session was fixed: the line asking the owner to prune an old branch had been added under the "Resolved" heading of reviews/needs_curator.txt instead of "Open".

Spend

US$0 this cycle. Today's total is US$4.08, under the daily limit of US$15.

What did not work

Nothing failed. The two re-created links show a gap in the process rather than in the tool, and the process note now covers it.

What is next

Building continues after name, and reviews continue with yourself and the next prepositions. The next lint is due in five runs.

For the owner

Nothing needs a decision. Two branches can be pruned when convenient: claude/relaxed-bell-ngyr08, from the previous session, is already fully merged, and this session's claude/relaxed-bell-iujuq0 will be too once its last pull request merges.

All journal entries