On March 12th, the Delphi team spent an evening going through the previous twenty-two days of Kalshi combo parlay data, over 1.28 million executed trades and $69 million in premium, trying to understand what kind of business Kalshi has been cooking up. In these twenty-two days, market-makers on Kalshi’s combo book collected $2.44 million in net margin against $69 million in premium.

A sportsbook’s parlay vig is around 15% in states like New Jersey, so Kalshi’s pricing is quite the boon for retail traders with the weighted maker margin around 3.53% (calculated from the figures in the previous paragraph). Kalshi’s edge is in the leg-level pricing, so when the leg events are multiplied, the compounding tilts into the maker’s favor by a few percent. And, unsurprisingly, 72% of the identifiable takers in our window lost money across their activity in Kalshi during this period.

Almost every single sport category was profitable for market-makers across those twenty-two days we investigated, except for NHL parlays, which cost the makers roughly 9% of their premium. As I was looking into the data, I found that most NHL parlays were profitable for makers, except for four-leg parlays. Four-leg parlays were running a NUTTY -96% margin on $191,000 of premium.

There was something fishy going on, so I dug a little deeper into which specific four-leg NHL combos lost the money. Of the $183,000 that market-makers lost on four-leg NHL parlays across the twenty-two days, $171,000, about ninety-four percent of the losses, came from a single ticker.

This ticker traded for only ninety-five minutes. It opened for fills on Saturday, February 28, 2026 at about 2:14 PM on the Eastern seaboard. And in the ninety-five minutes of its lifetime, it took 142 fills. The fills landed in sizes between thirty-nine and seven hundred twenty contracts, almost all of them at a price of ten to fifteen cents per contract, with 90 of the 142 at ten cents exactly. When the ticker’s games finished on the following Monday night, the combo paid out $1.00 per contract for 201,921 contracts. The maker had collected $25,530 in premium. The maker paid out $201,921.

The NHL averages about 6.2 combined goals per game, so 9.5 is a ridiculously high line. By rough league base rate, each “under 9.5” happens something like 90% of the time. Four of them on the same night, assuming independence, comes out to `0.90⁴ = 0.66`. Even accounting for correlation goal totals on a given night move together for all sorts of mundane reasons, from officiating style to post-Olympic conditioning to how often teams are playing on the second night of a back-to-back, the fair combo price was probably a little higher than that, so maybe $0.70-$0.85.

All the while, some market maker’s combo engine was selling it for a ridiculous $0.10.

The fills are quite intriguing. Looking at the timestamps: 19:14, 19:16, 19:45, 19:55, 19:59, 19:59 (a second order in the same second), 20:00, 20:04, 20:09, 20:13, 20:14, 20:14, 20:15, 20:18, 20:43 (three orders inside that minute), 20:44 (two inside that minute), 20:45, 20:47, 20:48, 20:49. All of the trade sizes are clustered: $500, $628, $720, $720 again, $720 yet again. For ninety-five minutes straight, the maker’s automated requoter kept re-posting the same size at the same price after each fill and whoever was on the other side kept lifting it.

The part that confused me at first was that the market actually priced each of the four underlying leg markets correctly. If you look up the individual over/under markets for those four games in Delphi Market’s historical API, each “over 9.5” contract is pinned one to five cents meaning the underlying leg markets had those games as heavy unders and priced them accordingly. But for some reason, a market maker happened to have a wildly wrong one.

I do not know the name of the market-maker nor the successful counterparty. It is still intriguing to observe these patterns happening over the RFQ data and to watch on the sidelines in this black box battle between the takers and the makers and speculate about what might be going on inside. My bet is that it had something to do with hockey resuming after the winter olympics hiatus. Whether there was a genuine technical glitch with a model or the traders at SIG and Jump were just sitting around their trading desks twiddling their thumbs during this period, we will never know…

The nature, origin, and limits of knowledge are notoriously difficult to assess. Some argue that truth is best approached through reason alone; others that only experience can tell us what is real; still others that knowledge is less discovered than constructed, by communities, institutions, and the slow accretion of consensus.

In the last few decades yet another tradition has emerged, one that advocates the use of markets to help resolve our difficult questions.

Delphi Markets has very kindly offered me their rostrum to speak to what happens when two such traditions disagree—in our case, about whether Israel is committing genocide in Gaza.

If you hear the term “genocide”, what comes to mind?

Well—if you’re a normal, worldly person these days—the answer is likely “Gaza.” Peer into the global consciousness and see—on much of the modern internet, “genocide” is chiefly used in the context of Israel.

A representative example of such discourse is this short-form video from Mehdi Hasan. You can watch it or read the transcript below, but I'll save you the trouble of either: like 75% is Hasan listing the names of genocide experts who support the charge, never with more than a sentence of introduction. Quoth:

Genocide experts are almost all in agreement that what’s happening is a genocide, per the legal definition under the Genocide Convention, acts committed with the intent to destroy in whole or in part a national, ethnic, racial or religious group.

So here are the names, the expert names you need to know, including Jewish and Israeli experts, on the subject of genocide to avoid being gaslit by apologists for Israel. First, there are the academic experts who say Gaza is a genocide.

Martin Shaw, who literally wrote the book, What is Genocide?. Melanie O’Brien, who is the literal president of the International Association of Genocide Scholars. Dirk Moses, who is the senior editor of the Journal of Genocide Research. Iva Vukusic, who helped prosecute the Srebrenica genocide.

William Schabas, who even defended Myanmar against charges of genocide, against the Rohingya, but says what Israel is doing in Gaza now is a genocide. There are also the Israeli genocide experts, the Israeli historians of the Holocaust, who say Gaza is a genocide. Professors Omer Bartov, Amos Goldberg, Daniel Blatman, Raz Segal, Lee Mordechai, Dr. Shmuel Lederman, who was originally opposed to calling Gaza a genocide, but now says, quote, what Israel is doing in Gaza is genocidal in every sense.

Then there are the human rights groups who say Gaza is a genocide. Amnesty International, Human Rights Watch, the legal clinics at top U.S. universities who say it’s a genocide, at Boston University School of Law, at Cornell Law School, at Yale Law School, the United Nations organizations and experts who say it’s a genocide, the U.N. Independent International Commission of Inquiry, the U.N. Special Committee to Investigate Israeli Practices, multiple U.N. special rapporteurs. In fact, speaking the other week to the Dutch newspaper NRC, the Israeli expert Raz Segal said, can I name someone whose work I respect who doesn’t consider it genocide? His answer, no.

The Israeli expert Shmuel Lederman says it’s now the consensus view among genocide researchers. So apologists for Israel and lazy and cowardly both sides media outlets might want to reject the charge of genocide in Gaza. They might want to pretend it’s hyperbole or even anti-Semitic.

But the experts, the actual experts on genocide, they know a genocide when they see one, and they see one in Gaza.

Now, I’m afraid I can’t say I’ve read Mr. Hasan’s book How to Win Every Argument. But from this video, I’d hazard a guess that he places significant emphasis on the appeal to authority—Hasan in this video finds it sufficient to make no other argument for his case beyond beyond a spread of nearly contextless names and institutions endorsing his conclusion.

But though the impression of this video is a bit exaggerated (there are some scholars who take other views, such as John Spencer), it’s certainly not too far off. It is indeed the relative consensus of professional “genocide experts”, in academia and NGOs, that the Israeli government intends to “destroy in whole or in part a national, ethnic, racial or religious group”—a la the Geneva Convention definition of genocide.

No less than ever-present Wikipedia—bound by a unique quasi-legal process to the consensus of “reliable sources”—currently reflects this view.

So the evidence here seems pretty strong; nigh unassailable. Could you, dear author, really make the argument that 12 experts, Yale Law, a U.N. Special Committee, even Wikipedia itself, is wrong! Clearly, you must be a Zionist1!

Well, dear reader, I could write an essay noting that Gaza's civilian casualty ratio—disputed, but likely not far from the 75% civilian implied by the Gazan Ministry of Health’s numbers—is within the normal range of modern urban warfare. After all—I could imagine writing—the US-backed campaign to retake Mosul and Raqqa from ISIS produced ratios of 2.5:1 or worse, and charges of genocide weren’t even contemplated. And though Israel has probably committed war crimes in several instances over the course of the conflict, basically every military action by Hamas—every single one!—violates the Geneva Convention: i.e. Hamas soldiers don’t wear uniforms, and basically only operate in civilian infrastructure which are both extremely serious war crimes.23

But I’m not going to do that. Instead, I will analyze a rather different sort of evidence.

The Striking Asymmetry Between Experts and Markets

Now, while the assessments of experts and institutions on questions genocidal are important for deciding, for instance, what it says on Wikipedia, they have little legal import. But there is an institution that does: The International Court of Justice, or ICJ, in The Hague.

The ICJ is the judicial organ of the United Nations, and the body with jurisdiction over the Genocide Convention itself. This is the institution internationally recognized to decide what is genocide, and what isn’t.

In December 2023, South Africa invoked that jurisdiction, filing a case against Israel under the Convention over its conduct in Gaza: The internationally recognized process by which genocide is officially determined is underway.

So what does the world think the ICJ will find? Mehdi Hasan—presumably—thinks the vaunted court will find Israel guilty, and if you were only watching his show, you’d get the impression that it’s a foregone conclusion.

But what if there were a different kind of knowledge production? Not the consensus of vibes, papers, essays—but a process that aggregates everyone with an opinion, and asks them not merely to assert it, but to stake something on it?

Well, you’re in luck! Previously banned by the CFTC, over the last five years such a process has become tenuously legal in the US: That process is the prediction market. Under this mechanism, one buys a share in an outcome at some price between zero and a dollar. If the outcome occurs, it pays out a dollar. If it doesn’t, it pays nothing.

I think it’s easy to underestimate the power and promise of this mechanism. The price at any given moment reflects the aggregate judgment of everyone willing to bet—and because being wrong is costly, the incentive is to be correct rather than merely to sound correct. A prediction market is thus—in Bryan Caplan’s terms—a tax on bullshit: It introduces a tangible cost for believing false things about the world.

So what do our prediction markets think about the genocide question? Here are two on the subject:

Surprised? Perhaps Hasan would be.

How should we think about this?

One anticipates the objection, “who cares what the ICJ thinks?” but this objection is harder to maintain than it might appear.

First of all: The ICJ is the only institution with the authority to rule that a state is committing genocide under the Genocide Convention.

What’s more, there is only one legal definition of genocide, and it’s the one Mehdi Hasan quotes in his video—“acts committed with the intent to destroy in whole or in part a national, ethnic, racial or religious group.” The ICJ uses that definition too; there isn’t an actual asymmetry. At the very least, the markets demonstrate the conventional definition of genocide, assessed by a rigorous process—ie, a legal one—pretty clearly does not fit.

More importantly, the UN’s court is obviously not biased in Israel’s favor. The UN General Assembly has adopted 173 resolutions against Israel since 2015, versus 80 against all other countries combined! The UN Human Rights Council has condemned Israel 112 times since 2006 — more than Syria, Iran, Russia, and North Korea combined. That the UN’s court is still very unlikely to find genocide should be understood as some sort of admission against interest.

Now—another objection. Manifold is a play-money prediction market; there are no real money markets on whether Israel will be found guilty of genocide by the ICJ (America is not yet ready, it seems, for markets on genocide).

It’s natural to be a bit skeptical of play-money markets, but the literature on the question is unanimous—empirically, it’s crystal clear that play-money markets like Manifold’s are as efficient as real money ones, which—as markets have a way of being—are very efficient indeed. One in five Manifold markets priced at 20% resolve yes (i.e., the prices are correct).

So this is the situation we are left with: experts on one side and markets, coupled to a legal process, on the other.

“Well, sure,” you may say, “you have ‘markets’ which you claim are ‘efficient’, and some acronym ICJ, but the other side has all those experts with names—they probably know something I don’t!”

How much you should trust experts relative to markets and courts sounds like a thorny epistemological question—and perhaps in some sense it is. Perhaps someone could write an essay extolling the many virtues of prediction markets: Professional pundits may have some incentive to get things right, but this incentive is dominated by their desire for experiences like appearing on TV, fitting in with their colleagues, and ideological self-expression. It’s only with markets—I could imagine someone writing—where there is a tangible financial incentive to get things right, and legal processes crafted by millennia of cultural evolution to acceptably resolve contentious disputes, that we have a real shot at making sense of a complex world.

But the ambitions of the intellectual tradition behind prediction markets are to transcend that. No, I do not intend to argue that markets are better at ascertaining the truth than experts. I intend to show you that this is the case.

Tetlock’s Razor

The anti-expert position, ironically, has an expert of its own: the eminent professor at the University of Pennsylvania, Philip Tetlock.

Tetlock, along with Robin Hanson, founded the field of forecasting. And he is most famous for his longitudinal surveys on the accuracy of expert judgment.

Between 1984 and 2003, Tetlock recruited 284 experts—academics and pundits—and asked them to make roughly 28,000 predictions about geopolitical and economic outcomes. Questions were drawn from the expert’s own areas of competence, so a Soviet specialist would be asked about Soviet affairs, and a trade economist about trade.

The results, published in Expert Political Judgment in 2005 after nearly two decades of questionnairing, were unambiguous. The average expert didn’t outperform random guessing, and performed significantly worse than simple statistical heuristics such as “extrapolate the current trend” or “assume no change.” They failed even to measurably outperform educated laymen.

This was, and remains, a very powerful critique. Quoth Richard Hanania’s magisterial 2021 essay “Tetlock and the Taliban”:

Phil Tetlock’s work on experts is one of those things that gets a lot of attention, but still manages to be underrated. In his 2005 Expert Political Judgment: How Good Is It? How Can We Know?, he found that the forecasting abilities of subject-matter experts were no better than educated laymen when it came to predicting geopolitical events and economic outcomes. As Bryan Caplan points out, we shouldn’t exaggerate the results here and provide too much fodder for populists; the questions asked were chosen for their difficulty, and the experts were being compared to laymen who nonetheless had met some threshold of education and competence.

At the same time, we shouldn’t put too little emphasis on the results either. They show that “expertise” as we understand it is largely fake. Should you listen to epidemiologists or economists when it comes to COVID-19? Conventional wisdom says “trust the experts.” The lesson of Tetlock (and the Afghanistan War), is that while you certainly shouldn’t be getting all your information from your uncle’s Facebook Wall, there is no reason to start with a strong prior that people with medical degrees know more than any intelligent person who honestly looks at the available data.

On contentious geopolitical questions, the weight you should give to the opinions of experts is—measurably4—around 0. The forecasters on Manifold are aware of Amnesty International, Yale Law School, and Martin Shaw. They have priced all of that in, and they have still arrived at 14%.

The world is full of such incongruities if you know where to look. The incentives of pundits are systematically bad: Lacking antibodies against ideological capture, they unrestrainedly proclaim whatever positions are socially advantageous. As Steven Pinker notes in Language, Cognition, and Human Nature: “People are embraced or condemned according to their beliefs, so one function of the mind may be to hold beliefs that bring the belief-holder the greatest number of allies, protectors, or disciples, rather than beliefs that are most likely to be true.” When rationalists say politics is the mind-killer, it’s because in this area of life the social rewards universally dominate the pragmatic rewards of more accurate world modeling; thus pontificators about politics have almost no incentive to get at the truth.

If you want to actually understand the world, don’t just read the New York Times—and certainly don’t just watch reels. The best exercise for your epistemic habits is to play the prediction markets: They give you structured practice, grade you, and incentivize you to understand why you were wrong.

I’ll leave you with one final such market. If you disagree with its assessment of the issue, don’t just let us know in the comments—go change the price yourself!

Subscribe now

The world was undone the last time a pandemic hit. A civilization that can price ten years of cash flows in a millisecond could not price, in time, the catastrophe under (and in) our noses.

Certainly, every piece of evidence we needed to predict COVID was available: a cluster of sick employees was noticed by an employer in China, an overrun waiting room by a charge nurse in Brazil, a supply chain tightening by a commodities trader in New York. Still, by the time it hit the press and political pundits, the pandemic was unstoppable. And what could we even do about it?

Erring on the side of caution is not ideal. Nobody has ever been promoted for buying ventilators a month too early. Nor does anybody want to explain a school closure that was unnecessary.

And although the CDC serves a much needed role, as a centralized agency, it is inherently slow. There is a long pipeline that information must travel: first, labs go from hospitals to regional/university centers who run the initial testing. This data is then pushed out to state epidemiology agencies who run their own risk models and do further testing. Finally, many days later, this information is forwarded to the CDC, who usually waits until Thursday, around 1 PM ET, to publish the official MMWR report.

I have been closely following the spread of measles and was curious whether anybody within this aforementioned chain had been front-running the Thursday CDC reports on Kalshi or Polymarket. I found that, on the Kalshi market, “Measles cases this year?” 49.3% of significant contracts (>$300) were traded on the two days leading up to the CDC report being released. I also pulled the trades from the analogous Polymarket event and across 71,071 fills, Polymarket’s yearly market has 47.7% of >$1,000 trades on Tuesday and Wednesday.

It is possible that this imbalance stills from state data getting released early. But, generally, the dashboards from states update uniformly randomly throughout the week, and so we would see a much more uniform graph if that was the case.

The Polymarket data has more granular monthly markets for measles too, which surprisingly do not have this pattern at all. For some reason, the distribution is flipped, with whale activity concentrating on late Thursday and Friday. The wallets that consistently make money on measles trade on Friday between 9 and 11 AM Eastern, often right at 10:00. Across fourteen Fridays in the dataset of monthly measles markets, ten had their peaks in that two-hour window, the sharpest on March 20, putting nearly 90% of its volume into it. It is probably safe to assume this is coming from traders buying the post-MMWR mean reversion once the news cycle has processed overnight and many square traders have taken positions.

It is not terrible even if there are potential insiders on these markets. The Polymarket measles market will never be able to replace an organization like the CDC. But it remains an incredible convenience to society to be able to aggregate insider opinion in a way that sidesteps formal channels and allows the public to view forecasts of the disease as quickly as possible and, on Polymarket specifically, to do so with trades that are timestamped, wallet-addressed, and visible on-chain to anyone curious enough to look.

One line that still seems vague is determining at what point disease markets become bioterrorism markets. CFTC 40.11 specifically prohibits contracts that involve terrorism and assassinations. And certainly if Kalshi can argue that the “Khamenei out” market a death market and halt it, then a similar argument can be made to show measles markets are bioterrorism markets. These markets are relatively illiquid and the lines on # of infections are wide, so it is not worth it for someone to superspread. In a future where a prediction market on a rare disease could get opened and gain a significant amount of liquidity, we should be terrified.