Editorial standards

Every publication says it checks its facts. Few show their work. Ours is visible in the articles themselves: each significant claim we report on is marked with a plain verdict, next to the evidence for it, and left open to revision when the evidence moves.

How verification works

A checked claim moves through three stages. First, extraction: our newsroom tooling surfaces checkable claims from launches, papers, filings, keynotes and reporting - the specific, falsifiable assertions worth checking. Second, assessment: a journalist gathers evidence, weighs it, consults primary sources and, where useful, attempts replication. Third, verdict: an editor assigns one of four statuses and writes the reasoning. No claim skips the human stage.

The four verdicts

We deliberately keep the vocabulary small. A verdict describes the state of the evidence today - not a permanent truth.

What counts as evidence

We weight sources roughly in this order: primary documentation and reproducible measurement first; independent expert analysis second; on-the-record statements third; and unverifiable assertion last. A vendor’s own benchmark is a claim, not evidence for itself. Where we can run a test - a proving benchmark, a reproduction of a reported result - we prefer to.

Evidence type	Weight	Example
Primary & reproducible	Highest	Protocol spec, on-chain data, a result we reproduced
Independent analysis	High	Third-party teardown, peer-reviewed paper
On-the-record statement	Moderate	Named engineer confirming a detail
Unverifiable assertion	Low	Keynote figure with no published basis

The role of AI

We use AI where it is genuinely good - reading everything, at all hours, and surfacing candidate claims faster than any newsroom could by hand. We do not use it to render verdicts. The judgement, the weighing of conflicting evidence and the words of every verdict are a journalist’s. Put plainly: AI extracts, humans decide. Being a publication about emerging technology, we hold our own automation to the same standard we hold others’.

Integrity & the open record

Verdicts change as evidence changes, and that is a feature. When a status moves, the previous state and the reason are recorded in the corrections log below - not quietly overwritten. Every article also publishes a SHA-256 fingerprint of its own text: if the words change, the fingerprint changes, and the edit is logged. The recipe is public so anyone can recompute it: SHA-256 of the article’s body text with HTML removed, entities decoded, all whitespace removed, and the result lower-cased. The history is part of the record.

Independence

Coverage decisions and verdicts are made independently of commercial relationships. Sponsors and event partners get no view of, or influence over, a claim’s status. Where a subject is also a partner, an advertiser, or otherwise connected to us, we disclose it in the piece. Read the full editorial-independence policy.

Corrections log

We will get some wrong. When we do, we correct openly and log it here - dated, categorised and never quietly deleted. We classify changes three ways: a factual correction fixes an error that doesn’t alter a conclusion; a clarification adds necessary context; and a verdict change moves a checked claim from one status to another. A changed verdict also stays visible where it was made: the article keeps the prior state and the reason. If you have evidence that bears on a verdict, send it to the desk - corrections that change a verdict are credited.