Start with the pages the business needs
List service, location and contact URLs before running a crawl. Identify which pages should be public and which should stay private or outside search. Compare that intent with response codes, robots directives, canonical links and the actual HTML a visitor receives. Treat an intentional preview noindex differently from a production mistake.
Repair the route, then verify the destination
A redirect is useful only when it takes the reader to a relevant working page. Check old URLs, internal links and navigation against the agreed route inventory. Test a missing page as well as a successful one; a friendly error page should still return an error status. Record the original behaviour before editing so the result can be compared.
Measure a real page on a small screen
Choose representative pages and repeat measurements under the same conditions. Separate downloaded JavaScript, image weight, layout movement and server response time. Prefer a small fix tied to an observed problem. A larger site audit score is not evidence that customers can now find an entrance, choose a service or finish a permitted action.
Leave an evidence trail the next person can use
For every change, keep the affected URL, before-and-after result, code version and test date. A release should match the tested artifact and include a usable rollback. When live visitor data is unavailable, label the gap instead of presenting a laboratory observation as customer experience.
Reference points
Policies can change. Check the current official guidance before changing a profile or publishing new location information.