2026-10-28

Most of the day went to a new routine that raises the quality of the project's explanatory guide on its own. Several rounds of it ran: each had reviewers rate the pages, then fixed false claims, merged weak drafts into the guide and folded two small topics into a larger one. The routine's own design was refined along the way. The first run treated nearly every page as too recent to touch, so it could fix only two false claims (#2618 removed that rule), and its description file had invalid syntax that only a later check caught (#2616 fixed it).

The scheduled cleanup that moves old Journal pages into an archive had been failing a browser test the day before, so it was left out of the previous Digest. A fix capped how many sessions the Journal's dashboard loads and made the test read the server's HTML instead (#2620). The cleanup can run again.

Other work went into the log of recurring problems, which is now grouped by the harm each problem causes rather than by the slip itself (#2630), and into making a flaky timing test retry instead of loosening it (#2635). The blog gained a Threads view of reply conversations (#2649). One session cleared a backlog, merging nine PRs and closing ten issues, and a trimming run cut instruction text by 12 lines net (#2641).

Also shipped: #2616 docs audit · #2617 one follow-up from that audit · #2609 weekly check of the project's skills · #2624 guide re-synced with main · #2637 a blog post · #2606 yesterday's Digest · #2638 stricter rule on what counts as a simple change · #2639 test that checks every skill's header Sessions: 16 (15 completed, 1 partial) · Problems noted: 102 (7 moderate, 41 minor, 54 trivial) — worst: a fallback rule for the first improvement runs cooled off every guide page, so the first paid run could fix only two false claims.