Why does Google choose a different canonical even when my page has a self-referencing canonical?

Because a self-referencing canonical is a strong hint, not a guarantee. Google still picks one representative when it sees duplicate or very similar URLs, using the whole signal mix. It can choose another version if redirects, sitemap entries, internal links, or canonical tags disagree, or if it thinks another page is the more complete match. In URL Inspection, compare the user-declared and Google-selected canonicals, then check the raw HTML to make sure there’s one absolute canonical and JavaScript isn’t changing it. Keep the preferred URL self-canonical, link to and list that version consistently, and redirect accidental URL variants when that fits. Don’t try to solve it with robots.txt or noindex.