Trust
How we count, in full.
The platform is only worth anything if the numbers are trustworthy. This page defines every figure shown on the site. If a definition here and a number elsewhere ever disagree, this page is what we will correct against.
Current figures
- Total active signatures
- 4
- Identity verified
- 0
- Email verified only
- 0
- Countries represented
- 3
Last recalculated 16 days ago.
Signatures are tied to the exact words that were signed
The headline number counts only people who signed the current version of the Accord. When the wording changes materially, everyone who signed the old text stays bound to the old text, with its own hash, and is reported separately as a legacy supporter.
We could quietly roll those people into the new total. Almost nobody would notice. We do not, because they agreed to different words, and a number built that way would be a lie however small.
We never count anyone else’s supporters
Other organisations have collected far more signatures than we have. Those people signed their appeal, not ours. Their counts appear on this site as context, each against its own source and the date we checked it, and they are never added to our number.
There is no code path that can combine them. The external figures live in a separate table that the counting worker does not read.
What counts as a signature
A signature is counted when a person has confirmed a verification code sent to their email address. Nothing before that point is counted. A form that was started and abandoned, an unconfirmed email, or a session that expired contributes nothing to any number on this site.
Active signatures
Completed, not withdrawn by the signer, and not invalidated by review. This is the number in the headline.
Verified signatures
Active signatures where the person completed identity verification (assurance level 2 or above). Today this number is zero, because identity verification is not built yet. We show it as zero rather than hiding the row.
Email verified
Active signatures where the email address was confirmed but identity was not independently verified. Right now that is every signature on the platform.
Countries represented
Countries with at least one active signature. Country is the signer’s stated country of primary residence. We do not treat IP geolocation as evidence of residence, and we do not infer it.
Professions and affiliations are self-declared
Signers can optionally say what they do and give a title and affiliation. We do not check any of it, and it is labelled as self-declared everywhere it appears. It never carries a verification badge and it never changes how a signature is counted.
It is shown publicly only for people who chose public display. If you signed privately, your background is stored for your own records and shown to nobody.
Verification levels
Not every signature carries the same assurance, and we do not describe them as if they do.
- Submitted, details entered, email not yet confirmed. Not counted.
- Email verified, a code sent to the address was entered correctly.
- Verified person, identity confirmed through an independent provider.
- Verified role or organization, a public office or a legal entity was confirmed against an authoritative source.
Duplicate handling
One active signature per person per Accord version. When a second attempt resolves to someone who already signed, we keep the first signature and return the original receipt rather than creating a competing record. One signal does this: a confirmed email address. There used to be a second, a fingerprint over name, country and date of birth, and it went when we stopped asking for the date. Name and country alone are not an identity, so nothing weaker replaced it: two people with the same name in the same country are two people, and treating them as one would refuse a real signature.
A shared IP address is never, on its own, a reason to reject a signature. Families, schools, offices, universities and public networks all produce many legitimate signatures from one address.
Withdrawals and invalidations
A signer can withdraw at any time. A withdrawn signature is removed from public search and subtracted from every total. The receipt ID stays valid and is never reissued.
Administrative invalidation is different: it happens only after review, always carries a reason code and evidence, and is written to an append-only audit log. Large batch invalidations require two people to approve.
How the totals are calculated
Signatures are written to a single transactional database, which is the only source of truth. Counters are maintained separately for speed. On a schedule we recount from the source of truth and compare, and when the two disagree, the difference is reported, not quietly overwritten.
The counter on the homepage can lag the database by up to a minute. It never runs ahead, and it is never inflated for effect.
What we do not do
- No illustrative, projected or placeholder numbers, anywhere, ever.
- No animated count-up that reaches the real figure slowly.
- No counting pending or unverified submissions in the headline number.
- No public date-of-birth search.
- No downloadable list of signers’ names.
- No selling, renting or sharing of signer data.
- No importing signatures from another petition, ever, under any circumstances.
- No adding another organisation’s count to ours.
- No recounting old signatures under new wording.
How the policy data is checked
The country and agreement pages are not a summary of the news. Each line is a single recorded fact, this actor, this event, this date, attached to the exact version of the official document it was read from, with the date it was read.
Nothing is published because a model produced it. A machine-drafted fact is marked as machine-drafted and cannot publish itself. A person has to open the source, read it, and put their name to it. Until that happens the fact exists in the database and appears nowhere on this site.
When two sources disagree, both are stored and the record says they disagree. We do not pick the one that reads better.
A country with no recorded action is shown as no verified action recorded, never as a failure. The difference matters: the first is a statement about our evidence, the second would be a judgement we have no standing to make.
The source is the fact check
Every policy fact on this site shows the document it came from, the date we read it, and the exact line it was read out of. Not a summary of the line. The line.
That is deliberate, and it is the most important sentence on this page. A badge saying we checked something is a claim you cannot test. A quotation next to a link is a claim you can test in about a second, without trusting us, a reviewer, or anyone else. Where those two disagree, the source wins, and if the source says something different from what we wrote, we are wrong and we want to hear about it.
We rank sources rather than treat them alike. Tier 1 is a primary authoritative record, a treaty register, a government publication. Tier 2 is an actor stating its own position. Tier 3 is independent reporting. Tier 4 is background. The tier is printed with every source, so you can weigh it yourself.
We do not read the news and summarise it. A claim here is one recorded event, attached to one document, with a date. If we cannot point at the document, we do not publish the claim.
Why we do not print the reviewer’s name
This platform is reviewed by one person. We do not print which of them checked a given fact, and we do not print their names. Reviewers are recorded internally and Terminatorfoundation answers for what is published; a private individual should not carry a public byline on statements about governments. Every claim links the source it came from, so a reader who doubts one can check it directly rather than taking a name on trust. If we have something wrong, tell us and the correction is published in the change ledger.
A published fact may note that a person checked it against the source. That note is a footnote, not the headline, the source above it is what you should read. It never names the person and never says how many looked. Reviewers are named in our database, and Terminatorfoundation answers publicly for everything here.
The reason is plain. A small number of private individuals check statements about governments. A name printed beside every one of those statements exposes a person in a way it does not expose an organisation, and it adds nothing a reader can use, you cannot verify a fact by reading a name.
What you can use is the source. Every claim links the document it came from and the date we read it. If you doubt something, open it.
If we have it wrong
Every page that publishes a fact carries a way to tell us it is wrong. You do not need an account, and you do not need to leave your name or an address, leave one only if you want a reply.
Sending a report changes nothing on this site. A person reads it and decides. If we agree, the correction is published in the change ledger with the same weight as the original, and the record that we were wrong stays there permanently. We never say who told us.
If we disagree, we write down why. A decision nobody can explain later is one nobody can defend later.
Known limitations
We would rather state these than have them found.
- This platform has one reviewer, so a published fact carries one person’s check rather than two independent ones. The Accord wording itself has not been through external expert review, and the page carrying it says so.
- Identity verification is not implemented yet, so no signature currently carries more than email-level assurance.
- Duplicate detection rests on the confirmed email address, so one person with two addresses can sign twice. Every appeal in this field that does not verify identity has the same gap.
- The national AI safety body dataset is not published, because it has not been researched and source-checked yet.
- External initiative counts are checked manually. They are date-stamped, so a stale figure is visible rather than silent, but they can be out of date between checks.
- Withdrawal removes a signature from every public total, but the daily history rows for days before the withdrawal are not retrospectively adjusted, those numbers were true on the day they were recorded.