Yezur Wiki:Daily routine

Task list for the automated daily maintenance routine (the Alompan project). It is report-first and read-only: each run surveys the wiki and prints a numbered report of issues and proposed fixes as its own output — it writes nothing to the wiki, and does not even log in. A human reviews the printed report and, on a separate run, has Alompan apply the chosen items.

On each run

Survey the wiki through the public API and print the report. No edits, no login. When unsure whether something qualifies, leave it out or mark it flag-only; a short, trustworthy report beats a noisy one.

1. Structural health

From the API (list=querypage):

Wanted categories (priority) — used but not created. Propose creating each with a best-guess parent and a one-line description; flag-only if the parent is unclear.
Wanted pages and templates, orphaned pages, uncategorised pages, double and broken redirects — list them, and propose the concrete fix where it is unambiguous.

2. Structured-data consistency

For each table in Special:CargoTables: rows missing a key field, duplicate values that should be unique, and references to pages that do not exist. Flag; propose a fix only where it is mechanical.

3. Markup

Linter errors (list=linterrors) — name the page and propose the fix.

4. Recent changes

There is no stored state, so use a rolling time window on list=recentchanges sized to the schedule (about a day for a daily run). For articles changed in that window: factual contradictions against linked or related canon; voice breaks (real-world references, out-of-frame intrusions, non-encyclopedic register — see Yezur Wiki:Conventions); and open threads (placeholders, dangling references) to propose as Talk-page notes.

The report

Print a dated, numbered list as the run's output, most severe first. Format each item as:

N. [category · severity] <page(s)> — the issue in a sentence. Fix: the concrete edit, or flag-only.

Close with a tally, e.g. "18 items — 3 high, 9 medium, 6 low".

Approving and applying

A human reads the printed report and picks the item numbers to apply. On a separate, human-triggered run, Alompan performs only those items — as ordinary encyclopedia edits, with in-character summaries. Items that need a genuine editorial or creative decision stay flag-only for a human.