Site Health Audit: the site audit tool that orders findings by evidence

It is the share of graded checks that passed, computed only from criteria the crawl could measure. A criterion whose required field the crawl did not capture is counted as needing evidence rather than as a pass, so the percentage never rises because something was not looked at.

By the severity and confidence encoded in the active NovaBrain criterion each finding belongs to. The fix-first queue starts with the most severe findings and opens on the page at fault, so the order reflects what the standard says matters, not which problem happens to occur on the most pages.

No. It shows the captured evidence, the criterion's guidance where supported, and an AI fix draft only when that action and a model are configured for your workspace. You apply changes on your site; the audit verifies them on the next crawl rather than assuming the draft was used.

By re-running the same criterion against a later crawl. When the fresh evidence is available the finding updates, the audit history keeps the earlier result beside the new one, and a regression is caught by the check that found the original problem. Nothing is closed by a checkbox.

The NovaBrain SEO and AEO criteria you switched on for the project: status and redirect health, canonicals and indexability, titles, descriptions and headings, structured data and answer readiness among them. The audit grades against that library and does not invent rules of its own.

It grades every URL the project crawl captured within your plan's URL limit, and the queue is built to be worked by severity rather than read end to end. Affected-page lists and exports keep large findings manageable, and the recheck closes them in batches on later crawls.