An investor put the question to me plainly: « Do we hold our own against 600 million publications? And what is the evidential value of the vetting? » An excellent question — because it contains the most widespread misconception about truth in the age of machines.
The misconception is to believe that veracity is measured by volume. That ten thousand sources asserting the same thing weigh ten thousand times more than a single one asserting the contrary. This is precisely how language models work: statistically, the most probable answer is the one most often written. The majority sets the norm — not the truth.
Veracity is not a vote. It is a qualification.
The history of science is a long list of majorities that were wrong. In Vienna in 1847, Ignaz Semmelweis imposed handwashing in a chlorine solution in his maternity ward: within months, the mortality of new mothers fell from roughly 12 % to under 3 % — one death in four spared, in the very ward where most women had been dying. Against the whole of the medicine of his day, which rejected his conclusions for decades. One voice against thousands. The voice was right, the vote was wrong. A system that counts occurrences would have buried Semmelweis; a system that qualifies sources would have singled him out.
To qualify is to ask of every claim four questions that volume ignores. Who asserts it — a peer-reviewed journal, a regulator, an anonymous blog? A source's credibility is not its audience. Who contradicts it — and with what standing? A qualified contradiction is worth more than a thousand lazy confirmations, because it shows where the real debate lies. When was it established — freshness: a figure from 2019 about a 2026 market is not false, it is stale, which is another way of no longer being true. In what context is it true — temporality and scope: « penetration reaches 72 % » may be exact for France and false for Europe, exact under one definition of online banking and false under another. A claim without its context is ammunition, not knowledge.
Credibility, agreement, freshness, context: these four facets do not add up into a vote — they compose into a judgment. And that judgment has a property a vote will never have: it can be explained. One can show why a claim holds, source by source, date by date — and why another, repeated ten thousand times, does not.
So do we hold our own against 600 million publications? The inverse question is the right one: faced with a corpus, what matters is not how many documents you hold, but how many you can qualify. An unqualified ocean weighs less than an instructed library.
This is the doctrine behind CRVV, the four-dimension confidence score that qualifies every claim in Evidence — credibility, relevance, verifiability, veracity — with freshness and context as explicit factors. See Evidence →