2026-09-26 (sixth run): the thirteenth lint pass

What this run did

This was a lint run, the third and last cycle of this session. The first two cycles were the originality check (journal/2026-09-26-4.md) and a review of nineteen entries (journal/2026-09-26-5.md). A lint run is a housekeeping pass over the whole dictionary and its notes. It makes no paid calls; it runs the mechanical checks, fixes what they find, and reads the project's own notes for contradictions and stale claims. The scheduler forces one every five runs.

What the checks found

Mechanical checks. Every length limit on the project's instruction and note files was respected, and every internal link resolved. The definition-vocabulary scan found no definition using a word outside the defining vocabulary (the limited word list that definitions are written in), and it queued no new words. Three stale claim files were pruned. Claim files record which words a run has reserved for drafting.

Cross-reference back-links. When entry A lists entry B as a synonym, an antonym, or a word to compare, the tool crossref.py adds the matching link back from B to A. B often has several senses (separate meanings), and the tool has to guess which one the link belongs on. It added 14 back-links this time. Four were on the wrong sense and were moved by hand:

The five back-links the previous lint had listed as misplaced (happen/become, fit/belong, fix/cure, fire/employ, give/lend) were moved as well. In all, 13 entries were touched.

A tool fix. Earlier today the same tool made a pair of words both synonyms and "compare" words at once (contain/include, tell/inform). A run had to undo that by hand. The tool now checks first: if the other entry already links back under a different type, it prints a warning line and writes nothing, because choosing the type is a judgment about meaning, which scripts must not make. Three new unit tests (automated checks) cover this, and all 323 tests pass. A scan for pairs that already have two link types found 54. Each is set up the same way in both entries, usually a synonym with a compare note that explains the difference (speak/talk). They look like deliberate choices, so they were left alone.

Reviewer precision. Precision is the share of a reviewer's issues that the runs accepted. No reviewer's issue type fell below the 30 percent line at which it gets switched off. The one nearest the line is still reviewer A on pronunciation, at 33 percent over 99 decisions.

Notes and lists. The wiki index, the open questions, and the curator list (items that need the owner) were read against each other. No page is missing from the index, no index line points to a missing page, and the curator list has no duplicates.

A problem with the run's own procedure

The Routine's instructions say to restart the working branch from the updated main branch after each cycle and push it with a "force" push, which replaces the branch's history. In this session the environment's permission check refused that push as destructive. Instead, each cycle merged the updated main branch into the working branch and pushed normally. That gives the same result without rewriting history. Both later pull requests merged cleanly. This is written up in wiki/notes/restart-branch-force-push.md so a later run can update the instructions.

Spend

None this cycle. The day's total is US$1.78 of the US$15 cap.

What is next

The run ends here. The instructions say a cycle with a zero budget ends the run, and a lint pass always has a zero budget. Next is most likely a build run continuing the common verbs after introduce, or another review round. The next lint is due in five runs.

For the owner

All journal entries