Technical site audit: how do I audit a large website without losing URL-level detail?

Every URL the crawler can reach within the project's scope, its robots and permission rules and the URL limit of your plan. The limit is not a sampling rate: up to it the set is complete, and the crawl records what the limit or the directives excluded, so the audit never presents part of a site as all of it.

Because status, metadata, canonicals, internal links, rendering, sitemaps and data quality are different decisions with different owners. One undifferentiated export mixes them and hides the severe among the trivial; a focused workbench asks one question of the crawl and lists the URLs that answer it, which is what a ticket needs.

Where rendering runs for a project, the crawl keeps the page after JavaScript beside the raw response and the rendering workbench shows what changed, so content or links that exist only after scripts execute are visible as such. Where rendering did not run, the render fields are marked unavailable rather than assumed.

The site audit orders findings by the severity and confidence encoded in the NovaBrain criterion each belongs to, not by how many pages a problem touches. The queue opens on the page at fault with its captured evidence, so the order reflects what the standard says matters and each item is a page you can open.

Yes. Every workbench filters to the affected URL set and exports the supported source fields, so a finding travels into a ticket as the list of pages and values behind it, with the crawl date attached. After the change ships, the next crawl re-runs the same criterion and the export can be regenerated to show what remains.

That it did not reach them. Private, blocked or limit-excluded pages are reported as not reached rather than inferred, and a crawl that stopped at a limit says so. The audit grades only what it captured, which is why its counts can be trusted for the scope it names.