[HosseinKarami]

Summary

Version 1.3.2. Fixes from the fourth full audit of the large WordPress multisite, with 1.3.1 and a short time budget. The run lost 17 AI answers, and a refresh changed many files with no new data. The list had 10 items. The details are in CHANGELOG.md.

The results from run to run

• A question whose pages the run did not test stays with its answers, as "not_checked": true. Before, the run deleted it. Each short check (refact-a11y check) deleted the questions of all the pages that it did not test. The report counts these questions apart, and the review tasks skip them.
• The pages of open issues of the sample-only tests (listener, focus, operate, template criteria, states) open first and join the sample. On the multisite, this adds 10 pages to a sample of 175.
• An own rule is "rule changed" only when its detection is newer than the engine that found the issue (src/own/versions.ts). Before, each release made every such issue "rule changed".
• The classes of a lazy load (lazyloaded, ls-is-cached, and others) are not part of the identity or the question hash. The old IDs move to the new ones with their status, answers, and solution.

The refresh

• It keeps replaces, the example of each question, and every part of the report. The audit writes the crawl parts of the report to runs/<run>/report-data.json, and pages.jsonl keeps the title, the reason of the sample, load_incomplete, and styles_missing.

The page loads

• A page whose stylesheet request failed loads once more. Chromium keeps an empty sheet for a failed stylesheet, so the engine watches the network. On the multisite, one unstyled render gave 5 false target issues.

The report

• A line for the pages that a budget left, by source and by section. The "Resolved" column counts this run only.

Test plan

• npm test, with a new test file, across-runs (19 checks). statuses now checks the "not checked" rule with the status code, because an audit adds such a page to the sample.
• The new test on the 1.3.1 code: 16 of its 19 checks fail there. The other 3 are a setup check and two checks that pass on both versions.
• A refresh with this branch on a copy of the multisite results of 1.3.1: no issue file changes, no answer is lost, and no example moves. 1,060 questions get the new item hash one time. Almost all of them hold a lazy-load class.
• A full audit of the multisite with 1.3.2.