In nearly every conversation about Veremet, there is a moment where someone asks: "So it's like a fact-checking site?"
The honest answer is no. And the reasons matter more than they sound.
Somewhere along the way of building this platform, we made a decision that surprised even us: we would not use the term "fact-check" anywhere in our product, our documentation, or our public writing. Not in copy. Not in headings. Not in conversation with reporters. The word would simply not appear.
This wasn't squeamishness about a controversial label. It was a structural decision. The vocabulary you choose for a verification system encodes assumptions about how truth works — who gets to decide it, what counts as a verdict, how confident anyone is allowed to be. The language of fact-checking encodes assumptions we no longer believe survive contact with the information environment we actually live in.
Here is what we mean.
The problem with "fact"
The English word "fact" is doing too much work. It is asked to carry both "water boils at 100°C at sea level" and "the policy reduced unemployment by 1.2 points." These are not the same kind of statement. The first is a settled empirical regularity. The second is a contested interpretation of a noisy dataset, almost always conditional on a model, a counterfactual, and a choice of comparison.
Most of what circulates as "information" today lives in the second category. It is not a fact to be checked. It is a claim to be examined. The grammar of fact-checking — binary, terminal, decided — flattens this distinction, and the flattening is where most of the damage happens. When a thirty-page report becomes a single icon — true, false, mixed — the work that went into the analysis is lost, the contested middle disappears, and the reader is left with a verdict instead of an understanding.
This is the first vocabulary substitution we made.
We don't check facts. We examine Claims.
A Claim is a testable proposition with an author, a context, and a chain of evidence behind it. It can be supported, challenged, qualified, or set aside as unresolvable. In most interesting cases it cannot be reduced to true or false without erasing what made it interesting in the first place. The word does the epistemic work the platform requires.
The problem with "checking"
The verb is worse than the noun.
"Checking" implies a single authority running a verification routine. The fact-checker reads the claim, consults the evidence, and issues a ruling. The ruling is the product. Everyone else is the audience.
This model has not aged well. Over the last decade, the phrase became a tribal signal — depending on the institution doing the checking, the same verdict was received as authoritative or dismissed as biased before it was even read. The structural problem wasn't the institutions themselves. Most of them were doing careful work. The problem was that the form of the output — a singular verdict issued by a singular authority — could not survive an information environment in which authority itself was contested.
Verification, done properly, is not a verdict. It is a structure of evidence with positions visible on it. People with different priors should be able to read it and see the same shape. They will weight the parts differently — that is fine, that is what humans do — but they should not be looking at different objects.
This is the second substitution.
We don't issue verdicts. We surface a Consensus Graph.
A Consensus Graph is a living structure: every piece of evidence visible, every source attributed, every supporting argument and every challenge in the same field of view. It updates as new evidence arrives. It records who weighted what, and why. A reader can disagree with the consensus and still see exactly how it was reached. That is the difference between a verdict and an inquiry.
The problem with "confirmed" and "contradicted"
Most fact-checking taxonomies have two columns: confirmed and contradicted. The verbs promise a finality the evidence rarely supports.
In our taxonomy, evidence is sorted into three buckets:
Supports. Evidence consistent with the Claim as stated.
Challenges. Evidence inconsistent with the Claim as stated.
Context. Evidence that doesn't directly support or challenge, but materially changes how the Claim should be read — caveats, scope conditions, history, definitions.
The third category is the one most fact-checking systems lack, and its absence is responsible for a startling share of the trust collapse around verification. Real claims have context. A study that "supports" a hypothesis at p=0.04 in a sample of 312 college sophomores supports it differently than a meta-analysis across forty countries. A statistic that is technically accurate within one fiscal year tells a different story across ten. Context is not noise to be edited out; it is part of the Claim's meaning. A verification system that has no place for it will, by its own structure, lie by omission.
The problem with "fact-checker"
Finally, the people. The label "fact-checker" suggests a role: someone employed by a publication to issue verdicts. It is a job title. The implicit credentialing is institutional — you are a fact-checker because your employer says so.
We wanted a word for someone whose authority comes from a different place. From domain expertise, demonstrated over time. From a track record visible to the public. From peers who can examine the work and disagree with it. From a stake in the outcome — including a financial stake — that makes their judgment costly to give carelessly.
The word we use is Maven.
A Maven on Veremet is a domain expert who contributes verification work in a public, attributable way. Their reasoning is visible. Their track record is visible. When their contributions are cited by paying users of the platform, they earn a share of the revenue their work generates. This is, as far as we can tell, the only verification model in which the experts doing the work have a continuing financial stake in the quality of it. The model is unfinished, and we are still learning what it does to incentives, but the underlying conviction is firm: domain expertise should not be a volunteer position, and verification should not be a salaried role at a publication that owns the output.
Mavens do not issue verdicts. They contribute to a Consensus Graph alongside other Mavens, AI analysis run across multiple labs, and (often) the original Claim's author. The consensus is what emerges from the structure. No single contributor controls it.
What "Community Trust" replaces
There is one more substitution worth naming. Most verification platforms have something called a "credibility score" attached to users, sources, or contributors. The word "credibility" implies an inherent property — you are credible or you are not — and the score is an institution's attempt to assign it.
We use Community Trust instead, and the difference is more than cosmetic. Community Trust is earned over time, in public, by visible work. It is not assigned. It is not awarded by Veremet. It is the residue of contributions other contributors found rigorous, that the Consensus Graph eventually endorsed, that held up when challenged. It can rise. It can fall. It is the only kind of trust we think is honest to display in a verification system, because it is the only kind a reader can themselves audit.
Why this matters
A new platform usually announces itself with new features. We are announcing ourselves, in part, with new vocabulary. The reason is not branding. It is that the existing vocabulary of fact-checking is part of what has made the public conversation about truth so brittle. Every term carries a model of how truth works, and the model embedded in "fact-check," "verdict," "credibility score," "confirmed," and "contradicted" is one that no longer survives the information environment we live in.
The model we are building has a different shape:
Claims, not facts. Supports, Challenges, and Context — not confirms and contradicts. A Consensus Graph, not a verdict. Mavens, not fact-checkers. Community Trust, not a credibility score.
This is not a softening of standards. The five-level scale on every Veremet Claim — Verified, Leaning True, Contested, Leaning False, False — does not let anyone off the hook. The strongest Claims still earn the strongest readings. But the middle is honored, the contested is named, and the unresolved is allowed to stay unresolved without anyone pretending otherwise.
The bet underneath all of this is that the public is ready for vocabulary that respects them. That readers can hold ambiguity. That a system showing its work is more trusted than a system issuing a stamp. That the right question to ask, when something is said in public, is not "who checked it?" but "what is the shape of the evidence?"
Veremet launches this summer. The first Founding Mavens are being recruited now. The first Claims are forming their Consensus Graphs.
None of them will be fact-checked.
They will be examined. Together.