The $FACY Truth Market: What If Fact-Checks Themselves Were Priced?
What ArAIstotle sees that the crowd doesn't for Prediction Markets
Right now, on a prediction market somewhere, the crowd is giving an event a 0.1% chance of happening. ArAIstotle is giving it 85%.
That gap of 85 percentage points of disagreement between an AI verifier and a thousand traders with real money on the line is the most interesting data point in prediction markets today. It's also live on our Truth Terminal, every minute of every day, on every major market we track.
This post is about what that gap means.
Markets and fact-checks are doing the same job
A fact-check is a snapshot. A market is a film.
A publication labels a claim "false" and that verdict freezes the moment the article goes live. New evidence may emerge tomorrow. Framing shifts next month. The headline judgment sits there, static, until someone with authority decides to revisit it.
A prediction market doesn't work that way. It keeps thinking. The price moves with every new piece of information, every new trader who shows up, every shift in what the crowd believes. The "verdict" is whatever the price says right now. And right now keeps changing.
Both are trying to do the same thing: produce a credible estimate of whether something is true. They just work on different time scales, with different incentive structures, and with very different failure modes.
Where the crowd breaks
The crowd is, on most things, pretty good. Prediction markets beat polls in 2024. They've out-predicted central banks on rate decisions, journalists on election calls, and analysts on company earnings.
But the crowd inherits the same problems that anyone reading the internet inherits. It absorbs viral narratives. It anchors on the loudest headline. And — most importantly — it treats every source as roughly equal weight, when in reality the gap between a well-reported piece and a poorly-reported one is enormous.
This is the part most market participants underestimate. A claim from one outlet is not the same as the same claim from another. And the difference doesn't break cleanly along the lines people assume.
What the Ad Fontes data actually shows
We integrate Ad Fontes Media's ratings into ArAIstotle's verification stack. Ad Fontes is an independent organization that scores news outlets on two axes: how reliable their reporting is, and how partisan it leans.
What the data shows, when you look at it properly, is that quality is not a property of the label.
Any outlet, on any side of the spectrum, can do journalism well — and any outlet, on any side, can do it badly. There are partisan publications that fact-check rigorously and centrist ones that don't. There are alternative outlets that break stories legacy media miss, and legacy outlets that publish viral nonsense. "Mainstream" versus "alternative," "left" versus "right" — these labels tell you almost nothing about whether a specific piece of reporting got the facts right.
The track record tells you a lot. The label tells you almost nothing.
So when the crowd reacts to a headline, it's usually reacting to that one headline without weighing how reliable the outlet behind it actually is, or how that outlet has performed on similar stories before. ArAIstotle weighs it. A high-reliability source carries more in the model than a low-reliability one — regardless of which side of the spectrum it sits on. Some of the largest deltas we see come from exactly this: a viral but poorly-sourced narrative driving market sentiment, while the verifier reads past it to the documents underneath.
This isn't about declaring some outlets good and others bad in some permanent way. It's about acknowledging that any source — across the entire spectrum — can be done well or done badly, and that the difference matters enormously when you're trying to price reality.
When the gap between ArAIstotle and the crowd is small, the crowd is probably right. When the gap is enormous, something is usually happening in the source-quality layer.
More on Ad Fontes Media: https://www.facticity.ai/post/ad-fontes-media-bias-chart
What we've built for Prediction Markets
The Prediction Market Monitoring layer on the Truth Terminal is exactly this: ArAIstotle's independent probability estimate sitting next to the crowd's price, on every major market, in real time. The delta is the signal.
We surface the top deltas — the markets where ArAIstotle and the crowd disagree most. We don't tell anyone to trade them. We surface them because the gap itself is information.
A few patterns we keep seeing:
The crowd under-prices binary near-term events that haven't entered the news cycle yet. Specific date-stamped resolutions, low-attention markets — ArAIstotle, working from primary sources, often sees 80%+ where the crowd is pricing under 5%.
The crowd over-extrapolates from recent momentum. When a stock or a candidate has had a great quarter, the market sometimes prices a much rarer outcome — will X be the largest company in the world by date Y — far higher than the underlying numbers support. ArAIstotle is more skeptical.
The crowd doesn't distinguish between "this should happen" and "this is on track to happen." Politically expected outcomes get over-priced; mechanically likely outcomes get under-priced.
These aren't trading recommendations. There are patterns in how attention prices reality.
The harder question
If markets and fact-checks are doing the same job from opposite directions, the obvious next move is to point them at each other.
The Terminal does half of this already: an AI verdict overlaid on a market price. The other half is the version of this idea we haven't built. What would it look like to run a market on a fact-check itself?
The instruments would be specific. Falsifiable. Things like:
Will this verdict be revised within six months?
Will ArAIstotle's confidence score on this claim cross 80% by year-end?
Will the primary cited source be retracted?
Each rewards the trader who saw something the original fact-check missed. Each generates a continuous price that tells you, in real time, how much the crowd trusts the verdict.
The fact-check doesn't disappear. It becomes the opening bid.
The honest part
We're not pretending this version is solved.
The oracle problem doesn't go away. Someone still has to resolve. Manipulation is real — low-liquidity markets get pushed by whales. And some people will find the idea of pricing truth distasteful, which we take seriously. Fact-checking already runs on incentives — reputation, traffic, institutional standing. A market just makes them legible.
Most claims don't deserve a market anyway. Liquidity is finite. The interesting design question is which claims earn one.
The bigger frame
Fact-checking, as it exists today, ships verdicts. Markets ship prices.
A verdict says: this is what we think. A price says: this is what we collectively, continuously, with skin in the game, are willing to bet.
Both have a place. But only one of them keeps updating after the article goes live.
The future of verification isn't a better verdict. It's a tighter feedback loop between the verdict and the price — faster, with more eyes on them, with consequences for being wrong. We've built one direction of that loop. The other direction is coming.
In the meantime, the Terminal is open. Watch the deltas. See where the crowd and the verifier disagree. That gap is the most honest map we have of where the work still is.
Thoughts on what a fact-check market should or shouldn't look like? Reach us at [email protected].
