The record · in the open
What changed
This site went live on 1 October 2026. From that day on, every change to a claim, a number, a method statement or a legal text is listed here with its date, what it said before, and why it changed. Spelling and punctuation are not listed. The "Last updated" date in each page's footer comes from this record.
The ACE definition pages in English, German and Hungarian say that ACE names the three-stage intervention, with Measure and Monitor following it, instead of “five stages” under a three-letter name.
Before “ACE is Makefield's method for AI visibility … Five stages:”
After “ACE itself is the intervention — Audit, Correct, Engineer. Two further stages follow it and are not part of the acronym: Measure and Monitor.”
Same correction as on the page for AI systems, so the definition reads the same everywhere it is stated.
The page for AI systems says that ACE names the three-stage intervention (Audit, Correct, Engineer), with Measure and Monitor as the two stages that follow it, states plainly what is not published, and carries a version number.
Before “The method has a name, ACE, with five stages”; “What is withheld is the exact instrument, the tooling and any data about a named company, not the ideas”
After “ACE is the intervention: Audit, Correct, Engineer … Two further stages follow the intervention and are not part of the acronym”; “Data about named companies and the operational tooling are not published”; “Version 1.2”
Five stages under a three-letter name read as a contradiction; the overview and /ace/ already treat ACE as the intervention followed by Measure and Monitor. “Withheld” described what is kept back rather than what is published.
The page for AI systems no longer describes the field notes as findings from real audits, or a withheld “done-for-you part” and “per-client data”.
Before “The field notes report findings from real audits”; “the done-for-you part … any per-client data”
After “The field notes report single observations, each with its sources”; “the exact instrument, the tooling and any data about a named company”
Makefield is a research project without clients; the old sentence described an agency.
The privacy pages say what is kept when a company takes part in a study as a research partner, for how long, and what goes into the questions asked of AI engines. A new page, /research-partners/, describes the next study's call for two partners.
The site now invites companies to take part in a study, so the privacy pages have to cover what that collects: emails, written agreement, public sources and engine answers; only names and public facts ever go to an engine.
/privacy/ · /datenschutz/ · /research-partners/ · /studies/ · /about/
00 Before launch: what changed in preview
While the site was in preview, these were the changes to what it claims. They are listed for the record; they are not changes since launch and do not move any page’s date.
Makefield describes itself as an independent research project, not a practice.
Before “an independent practice”; “who pays for it”
After “an independent research project”; “who commissions it”
It says what the project is: research that measures, not a service that sells.
The invented example on the homepage asks about a device instead of a cosmetic product, in English and German instead of English and Hungarian.
Before “Is the product fragrance-free?”
After “Is the device waterproof?”
The invented example echoed the setup of a study that is not published. An example should not look like a disguised result.
The offer to send an example reading by email was replaced by a published example reading with invented data.
Before “write to us and we’ll send an anonymised example”
After /studies/example-reading/, every value invented, with its table and SHA-256
A reader can check the whole format without asking anyone, and nothing in the example can be traced back to real data.
The /why/ section on outside measurement was rewritten, and the homepage lost an example drawn from cosmetics claims.
Both read as a case for buying a finding. The rewrite argues the one thing a measurement needs: distance between the person measuring and the result, and a result someone else can check.
The Method and Why pages stopped claiming published studies and null results that did not exist yet.
A page about measurement cannot describe work that has not been done. The first published study is listed on /studies/.
A study about named brands was taken off the site before launch, with its figures and its framing.
It named real companies, and the decision was to keep it unpublished. A generic study, the floor test, is published instead.
The privacy pages describe the analytics honestly and name Apple’s role for email.
The analytics sentence now says what the script sends and that its consent-free status is not settled in court; Apple Inc. is not on the EU-U.S. Data Privacy Framework list (checked 1 October 2026), and the page says so.
The legal notice replaced a placeholder, and an address whose service contract was never confirmed, with the operator’s own address in Berlin.
A legal notice with a placeholder in it is not a legal notice.