Three copies of every page and one read
A rival’s sitemap listed 1,611 addresses. They were 558 pages in three locale folders. Reading all three triples the work and makes a page count a lie.
The first thing a whole-site read has to decide is what a page is. The rival we read on Sunday publishes a sitemap of 1,611 addresses, which sounds like a large site. Sorted by path it is three copies of a medium one: 550 addresses under a UK folder, 550 under a US folder, and 511 with no folder at all. The bodies under the two country folders are the same pages with the spelling and the currency swapped, and the bare set is mostly the same pages again from before the folders existed.
Picking one copy
We took the UK set, because our readers are UK food halls and because the UK pages are the ones that mention UK partners and prices. Then we added the bare pages that had no UK twin, which was nine of them, so that a page that only exists outside the folders was not missed. That made 558 addresses to fetch instead of 1,611, and it is the number that goes beside the date on the comparison page.
The reason to be strict about this is not the fetch time, though cutting a run by two thirds is welcome when the pages are read in parallel six at a time. It is that a page count is a claim. “The word kiosk appears on none of 1,611 pages” sounds three times as strong as “on none of 558”, and it is the same evidence counted three times. A reader who later notices the folders would be right to ask what else was counted three times.
What stays true across the copies
A locale copy can differ in one thing that matters: a feature sold in one country and not another. We checked for that on the cells that moved. The sentence that moved their money-split cell from blank to tick is on the UK multi-vendor page, and the US copy carries the same sentence. The kitchen-screen routing sentence is on both too. If a sentence had been on the US page only, the cell would have stayed blank for a UK reader, with a note saying where the sentence lives.
The last trap is the sitemap’s modified date, which stamps every copy with the same day whether or not the words changed, so it cannot tell you which of the three copies is current. We had learned that on another rival’s site already, and the rule carried over: read the words, pick one copy on purpose, and write down which one.