Markets Find Prices, Not Truth

The world is full of people who are famous for being confident but terrible at being right. A pundit predicts catastrophe, is wrong, suffers no consequence. A hedge fund manager calls the top, misses by two years, gets a book deal. We live in an attention economy that rewards extremity over accuracy. Prediction markets promise to fix this. $44 billion flowed through Kalshi and Polymarket in 2025. Polymarket's CEO calls prediction markets "the most accurate thing we have as mankind."

The thesis is seductive: put your money where your mouth is, truth emerges. Replace the wisdom of credentialed bloviators with the wisdom of crowds who have something to lose. I've even built a little tool based on this premise. A Polymarket tracker that surfaces whale trades and front-runs news. Follow what smart money does, not what pundits say, blah blah.

Here's the problem. Markets can only price questions that settle cleanly. Will the Fed cut rates? Binary, time-bound, resolvable. Will this policy be good? Will this technology prove dangerous? These questions have no settlement date. They cannot be priced. They are the questions that matter most, but you cannot price what you cannot settle.

Prediction markets' accuracy comes from their constraint to tractable questions. When Polymarket outperformed polls on election night, this told us something about election forecasting. It told us nothing about the health of the Republic. You can price Biden dropping out. You cannot price the reasons and whether they represent democratic failure or renewal.

We are building an epistemology that systematically privileges what can be bet on over what matters. The financialization of truth is not fixing our thinking, it just reshapes where we put our attention.

The Goodhart Trap

Famously, when you create a metric for truth the metric becomes the target. When the target is hit, the original purpose tends to be lost.

Tetlock's superforecasters succeeded partly because they were unencumbered by professional reputation. They had, in his words, "little ego invested in each forecast." They were amateurs, updating beliefs carefully without career consequences. Scale the tracking, formalize the scoring, attach it to status and money, and you reintroduce exactly the dynamics that made credentialed experts unreliable in the first place. A cottage industry of Brier score optimization emerges. The system meant to surface truth instead surfaces those who optimize for the metric.

I built a tool to track which press personalities get predictions right under the accountability thesis. But if accountability become widespread the equilibrium response is predictable. Pundits make fewer falsifiable claims. Or they make more claims, distributed across enough topics that regression to the mean protects them. Or they retreat to podcasts where the record is harder to audit.

Accountability systems select for those who game accountability systems.

Being Early Is Not Being Right

My whale tracker reveals who moves markets. A wallet places $50,000 on a geopolitical outcome. The price shifts. News correlates. The signal seems actionable.

But moving markets and understanding reality are different things. A trader with insider information can move prices without any superior insight into underlying dynamics. A trader with deep structural understanding but poor timing gets wiped out before their thesis materializes. Markets reward being early. Being wise is a different skill.

This matters because we are building an epistemology that treats market prices as proxies for truth. If Polymarket says 73%, that becomes the credible estimate, the prior against which to update. But prices reflect capital and conviction among current participants, not reality. In liquid markets with diverse traders, this approximates dispersed knowledge. In thinner markets, it reflects whoever happens to be trading. One well-capitalized individual with a strong opinion can move prediction markets. We have seen this repeatedly. The resulting price gets reported as "the market" speaking, as though collective intelligence rather than one person's position.

The Outsourcing of Judgment

We are building systems to tell us whom to trust because trusting the wrong people has become expensive. Misinformation, confident wrongness, motivated reasoning. The costs are obvious. The solution seems obvious too: automate the evaluation. Build a credibility layer. Let markets sort the reliable from the unreliable.

But deciding whom to trust is itself a form of thinking. It requires engaging with arguments, weighing evidence, noticing inconsistency. We are trying to outsource this labor.

The promise: you do not need to evaluate whether a pundit's argument is sound. Check their track record. You do not need to form your own probability estimate. Defer to the market. The score does it for you.

This is attractive because it is efficient. It is corrosive because it atrophies the very capacities it claims to enhance. If you never practice evaluating arguments, you lose the ability to evaluate arguments. You become dependent on the scoring system. You become worse at thinking.

The drive for accountability is a byproduct of our collective rejection of expert deference. The implication is that we should think for ourselves. But if the response is to build new systems of deference that tell us which experts to trust based on algorithmic scoring, we have replaced one priesthood with another.

What Cannot Be Priced

The questions that matter most are the ones prediction markets handle worst. Questions about meaning, trajectory, consequence. Questions where the framing itself is contested. Questions where being right depends on timescales that exceed market settlement.

Was the Iraq War a mistake? A prediction market in 2003 could price whether Baghdad would fall. It could not price whether the invasion was wise. We have outcomes. We have costs. But answering requires agreement on what counts as success, what counterfactuals are relevant, what time horizon matters. These cannot be specified cleanly enough to bet on.

Is AI development proceeding safely? You cannot settle this bet until consequences unfold. You cannot price existential risk without specifying trigger conditions, and the most concerning scenarios are precisely those where the trigger is crossed gradually, ambiguously, in ways that leave room for interpretation.

The financialization of epistemology privileges tractable questions over important ones. This is not a bug. It is intrinsic to how markets work. The worry is that our collective attention follows the prices. We learn to care about what can be bet on. We learn to frame our uncertainties as prediction markets would frame them. We become better at forecasters and worse at thinkers.

The Map and the Territory

Prediction markets aggregate information. They do this well, better than most alternatives for questions they can address. The ability to see, in real time, how capital is being allocated across possible futures provides a check on narrative capture. It humiliates confident wrongness, at least eventually. These are not small things.

But markets find prices. They do not find truth. The distinction matters because price and truth diverge precisely when truth is most valuable. When everyone agrees, the price is correct and trivial. When capital and insight are misaligned, the price is wrong. When the question cannot be specified cleanly, there is no price at all.

The risk is that we mistake the map for the territory. That we treat market probabilities as the probabilities, track records as wisdom, the aggregation of bets as the aggregation of understanding. That we build the machinery of epistemic accountability and find, years later, that we have automated the appearance of good judgment while the substance has quietly decayed.

My vibe coded projects operate on the premises I question. Tracking whale trades assumes whale behavior is informative. Scoring pundit accuracy assumes accuracy is what matters. These assumptions are probably more right than wrong. But "more right than wrong" is not the same as true, and the difference matters when you are building systems that shape how societies decide what to believe.

Thinking is hard. There are no shortcuts. Markets help. Accountability helps. Neither substitutes for engaging with arguments, weighing evidence, noticing when you are wrong. The promise of prediction markets is that you can outsource this labor to the crowd. The reality may be that what you outsource, you lose.

The crowd is often right. It is never you.