In both modes every counted folder row nests a preview of its own DIRECT child URLs in urls (capped at urlsPerFolder, with the uncapped total in matchedUrlCount), so one request returns a fully renderable tree level -- folders and the pages inside them -- without a follow-up call to folders/urls. That endpoint remains the way to page through a folder whose preview was truncated; it lists in the same default (URL, id) order, so the preview's URLs reappear there in the same relative order. It lists every direct URL of the folder without this endpoint's view filtering, so its rows and totals can exceed the preview's matchedUrlCount (e.g. on censuses with IGNORE patterns or in excluded views).

Browse mode (no searchPattern): With host, one level of that host's folder tree; without host, the root folder of every host on the census. Every row is a counted row, so every row carries the preview: all of its in-view direct URLs, up to the cap. Counts always cover each folder's entire subtree, whether they are read from the persisted folder rollup or computed live, so a folder's totalUrls does not change when the census finalizes.

Search mode (searchPattern supplied): A folder is a hit when its OWN name (folderName; a host's root folder is named "/" and is instead findable by its hostname) contains the pattern, or when it DIRECTLY contains a URL whose own page name -- the last segment of its path, not the folders above it -- contains the pattern. Hits come back in folders with matched=true, each nesting its matching direct URLs in urls. Neither kind of match inherits: searching "shoes" does not return every page under a /shoes/ folder (the folder is a name hit, but its urls preview lists only direct children whose own name matches), and a page named shoes.html puts its containing folder on the page without making that folder's name a match.

folders additionally carries the unmatched ancestor folders needed to render a connected tree path, with matched=false and no urls/matchedUrlCount. Every hit is connected back to the root (or to parentFolderPath when supplied). parentFolderPath still requires host, but host on its own is a whole-host search.

Pagination counts matched folders only: totalCount is the number of matching folders, and one page is pageSize matched rows. Nested urls entries and ancestor context rows are additional and never counted. A matching URL appears exactly once across the whole pagination -- nested under its containing folder, on the page where that folder sorts.

Where the counts come from: the excluded view (returnExcludedOnly=true) is always computed live from URLs, and so is the included view whenever the census is running, has IGNORE patterns configured, or has not finished folder aggregation; otherwise the included view is served from the persisted folder rollup. A search result's folder counts always describe the folder's whole subtree on both paths -- they are never restricted to matching URLs -- so the same folder reports the same totalUrls whether it was reached by browsing or by searching. Persisted counts still bucket status codes slightly differently from live ones: live treats any status other than 0 as visited, while the persisted unknown-status bucket additionally absorbs missing, 1xx and 6xx+ codes and reports them under urlsToVisit. visitedUrls + urlsToVisit always equals totalUrls on both paths.