On February 28, 2026, US and Israeli forces struck Iran. A trader identified as Magamyman won $553,000 on Polymarket betting that Khamenei was dead. Total Iran bets across Polymarket hit $529 million, with new wallets appearing just before the strikes with position sizing that couldn't be explained by public information. Bloomberg noted the pattern. NPR reported it. The implication was obvious.
The market worked. The insiders were right. The price was accurate. And the controversy about whether this was legitimate obscures something more interesting: what this case actually reveals about prediction markets as a forecasting tool.
The standard case for prediction markets rests on distributed knowledge. Thousands of independent analysts, each holding a small fragment of information — a satellite image, a leaked memo, an unverified tip, a pattern in shipping data — trade their private estimates through price. The mechanism aggregates these fragments better than any central analyst could, because the aggregation happens through profit incentive rather than institutional hierarchy. It's a correct argument, on a large enough class of problems.
The key word is "distributed." The theory requires that the relevant information is spread across many people, no single one of whom has a decisive advantage. Elections, economic forecasts, sports outcomes, commodity prices — these fit the model. The information is imperfect everywhere, marginally better in some places, but nobody has the answer.
Military operations are different. The decision to strike Iran on February 28 was made by a small number of people in classified settings. The planning documents, the NSC discussions, the mission architecture, the Israeli coordination — none of this was distributed. It was concentrated. The people with access were not a large, diverse population running independent models. They were, at most, hundreds of individuals with compartmentalized clearances.
Some of those people, or people in their information networks, were apparently trading Polymarket. This is what the position sizing suggests. Not that insiders went directly to the market — that would be obvious and prosecutable. But that information, like water, found paths: conversations, inferences, relationships between people who could piece together a fuller picture than any single cleared individual.
Here is the paradox: the Iran prediction market was more accurate because of the insiders. Their information was real. The strikes happened. The price moved before the public signal because private signals were flowing into the market. In this sense, what looks like corruption — trading on non-public information — is also the mechanism by which the market delivered its most impressive accuracy.
Prediction market advocates want to claim this as evidence that the mechanism works. It did work. The price found the truth before the news did. But the mechanism that found the truth is not the mechanism that was advertised. It wasn't distributed independent analysis by informed participants. It was a small number of people with access to classified information expressing that information through price.
The market was a vehicle for information transfer, not information aggregation. Those are different things.
The practical problem is interpretation. When you look at a prediction market price, you can't determine which regime you're in:
In one regime, the price reflects 1,000 analysts independently running open-source models. Your model is one more data point. You can assess whether you have information that isn't priced, and if you do, you can update the market with it. Your analysis is additive.
In the other regime, the price reflects 3 people who read the classified briefing, plus 997 people responding to a price that those 3 already moved. In this regime, the price is more accurate than anything you could produce from public information. But your analysis is worthless — not because it's wrong, but because the price has already incorporated the answer. You're the 998th person to analyze a question that was solved before you opened your browser.
The clearance gap means you can't know which regime you're in. The price looks the same from the outside.
The gap is widest for events with information structures that reward concentration over distribution: military operations, regulatory approvals, central bank rate decisions, M&A. For these, the decisive information is held by a small number of people, and the prediction market price is most likely to already encode that information through leakage.
The gap is smallest for events with genuinely distributed information: elections where millions of voters haven't decided, economic forecasts where the relevant data is public, commodity prices where supply and demand data is widely accessible. Here the distributed-knowledge model actually applies, and your analysis can genuinely add to the market.
For events in the concentrated-information category, the right use of prediction markets is not as a probability to reason about and update. It's as a binary sensor. When a price moves unexpectedly — especially in a low-volume market, especially from new wallets, especially before any public signal — something private is flowing in. You can't know what. But you know that someone who does know something is expressing it. The market tells you a private signal exists. It cannot tell you what that signal is.
Magamwyn made $553,000. The market reached $529 million in Iran bets. The price was accurate before the news. None of this is an argument against prediction markets — it's an argument for understanding their actual operating mechanism. They aggregate information. They don't filter for how that information was obtained. The price is the best available signal, conditioned on whoever happened to have information and chose to trade it.
For geopolitical surprises specifically, the best available signal often comes from people with access you don't have. The market captures it. You see the price but not the epistemics behind it. And for the question of whether you can add anything — whether your own model, your own analysis, your own read of the evidence contributes anything — the honest answer is: only if the information is distributed, and only if you have something the price hasn't already incorporated.
The Iran case was an answer to a question that was already solved by the time most traders asked it. The clearance gap was total. The market worked perfectly, and for most participants, it wasn't useful at all.