Emerging technology runs on claims. A model is state of the art. A protocol is orders of magnitude faster. A robot achieves human-level dexterity. A token will change everything. Most of these assertions are repeated long before anyone checks them, and by the time the checking happens the narrative has already moved on. This is the problem we were built to address.
The claims problem
The technologies we cover - artificial intelligence, blockchain, zero-knowledge, robotics - are unusually hard to verify and unusually easy to hype. They are technical enough that few readers can evaluate the claims directly, fast-moving enough that scrutiny always lags, and consequential enough that being wrong matters. The result is a discourse where the loudest assertion often wins, and where “verifiable” has become a marketing adjective rather than a property anyone confirms.
We take our method from the field we grew out of: a culture whose default is to check rather than to take on faith. That habit built the tools this whole industry now depends on, and we think it applies to journalism about the industry just as well.
Verification as the product
Every publication says it checks its facts. Few show the work. Our answer is to make verification a visible, standing thing rather than a private habit. That is what our claim-checking is: to show the work where it happens: every article carries its sources, and the checkable claims inside it are marked with a plain verdict - verified, disputed, refuted - next to the evidence for it. When we get one wrong, the correction is public.
Trust, for us, is not a tone. It is an artefact we hand you - and one you can check without our permission.
It is also why our own work carries its evidence with it. Articles list their sources. Checkable claims are marked with a verdict and their source, right where they are made. Where we use automation - and we do, because reading everything is exactly what machines are good at - the judgement stays human. AI surfaces claims; editors decide verdicts. We hold our own tooling to the standard we hold everyone else’s.
What that commits us to
Two things, mostly. First, we will publish less than we could, because verification is slower than assertion, and we would rather be right and late than fast and wrong. Second, we will be visibly wrong sometimes, and when we are, the correction will be public and logged rather than quietly edited away. A verification brand that hid its mistakes would be a contradiction.
The wager is simple: that in a moment when synthetic media has made everyone unsure what to believe, the most valuable thing a technology publication can offer is not the fastest take but the most checkable one. That is what we are for.
What verification cannot do
Honesty about method includes its limits. Verification settles questions of fact with public evidence; it does not settle values, and we will not pretend otherwise. Whether a surveillance system works is checkable; whether it should exist is a judgement, and when we make one we will label it as opinion rather than dress it in a verdict. Predictions get the same treatment: no evidence exists about the future, so forecasts here are arguments to be weighed, marked as such, never rated. And some true claims are unverifiable on public evidence - private figures, unreleased systems - in which case the accurate verdict is not a guess but an honest “cannot yet be checked”. A verification brand dies the day it stamps certainty on things it cannot see; the discipline of saying “we don’t know” is the product.
Hold us to all of it - the corrections log is public, and the fingerprint on this page is recomputable by anyone. That is the deal.
- THE VERIFIER
THE FOUNDING DESK
