Case: 12,000 non-canonical URLs removed from sitemap

Case: 12,000 non-canonical URLs removed from sitemap
A store listed both ?color= variants and their canonicals. Google kept choosing the variant, splitting signals.
What we did:
— Checked each sitemap URL's rel=canonical
— Found 12,000 URLs pointing their canonical elsewhere
— Kept only self-canonical URLs in generation
Measured outcome over 6 weeks:
— 'Duplicate, Google chose different canonical': down 81%
— Correct canonical selected on 96% of audited products
Checklist:
✅ Every sitemap URL is self-canonical
✅ No parameterized duplicate appears
✅ Canonical target equals the listed URL exactly
Definition of done: every URL in the sitemap has a canonical equal to itself.