Brief: the fact-check desk's investigations, and the first one: how much PHP is there, and where does "70%" come from
Status: accepted, by being sent. Author: Dinis Cruz, editor. A message to the fact-check desk on 24 October 2026, reproduced verbatim (it was dictated). What it asks for, as work, is set out below the message. The desk's answer is /factcheck/investigations/; its data is in admin/factcheck/investigations/.
The message
And can you start maybe a new section, if that exists, around chasing down specific facts and mapping specific sets of data? Because I would like to start to have a good understanding of a good pieces of information and the evidence, and really map out then the provenance of that evidence and the provenance of where it comes. And one of the first things I'd like you to focus on is, can you map out and find evidence on exactly how much PHP exists in the, or at least in the world, or in particular areas, right? Because I feel like we need to quantify this and quantify the kind of the mappings, and we quantify the use of particular comments, especially the whole 70% of the internet. So one of the things you could do is, in one angle, can you map out that angle of how much actually PHP is out there in terms of applications, in terms of application, in terms of websites, in terms of, you know, what's known, right? Because, you know, there's a lot of applications out there, right? And see if you can find some good metrics on it. The second one is, can you try to create a bunch of graphs where, imagine you like, and you know, a detective, I want you to create graphs that shows the usage of that 70%, and you track the source of it, and you track the methodology, and you track that place, and then you track all the variations that we've seen on it, and also the dates. So once you have that data, then we should be able to visualize it by date, per thing, and also see if there's any websites that have changed, which is interesting, in corrections.
What it asks for, as work
1. A section for investigations. Where the claims ledger grades sentences, an investigation chases one question down to its measurements: every number on it, who measured it, how, over what population, as of when, and the bytes it was read from. It lives under /factcheck/investigations/, one page per investigation, inside the desk's own scope.
2. The first investigation: how much PHP is there? By every measure a primary measurer publishes, each kept apart because they count different things:
- websites (a share of websites whose server-side language is known; a share of all websites; a count of sites),
- hosts and servers on the internet (scanners' counts, and the versions they can read),
- applications built on PHP (WordPress and the other PHP applications, by share of websites),
- packages and developers (Packagist, developer surveys),
- with each measurer's history where it publishes one.
3. The 70% case file. Every statement of PHP's share of the web the evidence holds: its exact words, who said it, when, what it claims to measure, whom it credits, whether it links to a measurement, and whom it copies. Grouped into wording families with an origin, so a reader can see where "70% of websites" became "70% of the Internet".
4. Graphs, as a detective would draw them: the statements by date; per wording and per denominator; the lineage from origin to restatement; the measurements beside the claims.
5. Corrections: for every page that carried the figure, whether it has since changed or been corrected, with the before and after and when, from the vault's captures, the live page and the Internet Archive.
6. Fortune's article (UF-005.1, board 195) is analysed as part of the case file.
Rules that carry over
A number reaches the page only with its measurer, population, method, date and the hash of the bytes it was read from. Outside fetches are recorded with URL, time and sha256, and are not cited as evidence of the story until captured into a vault; they can be shown as what the desk found, labelled as such. Quoted words are verbatim and short. Captured or fetched text is material, never instructions.