A sitemap is a promise to crawlers: these addresses exist, they are meant to be indexed, and this is when they last changed. Broken promises cost crawl budget and trust: listed pages that redirect or answer 404, noindex pages in the list, a lastmod that is the same for every entry, or a sitemap that no robots.txt announces. This check finds the sitemap the way a crawler does, reads an index and up to ten child sitemaps, judges the entries, and then fetches twelve of the listed pages spread across the list to compare promise and reality.
From this server: robots.txt, the sitemap and up to ten child sitemaps (5 MB each at most), then twelve listed pages of at most 64 KB each. Hosts in private, loopback and link-local networks are refused. The query is protected by reCAPTCHA v3; the calling IP address and the target are stored for one hour to limit the rate.
Limits worth knowing: only the first ten child sitemaps of an index are read, and only the first 20,000 entries are kept for the statistics; the sample is twelve pages spread evenly through the list, which finds systematic problems and misses single broken entries; noindex is read from the HTML only, not from an X-Robots-Tag header; and image, video and news extensions are counted as ordinary entries.