They collected all of it during the same twelve months in which the evidence for their founding claim came apart in public. The claim is that crowds price the future better than experts. The record says a few percent of traders do the pricing, the most-watched markets are the least accurate, and the venue that won did so by selling sports.
Our new report, out today, is about the distance between those two things. Here is the short version.
Kalshi won the category, and it won on sports
The leadership question is settled, and most coverage still has it backwards. Polymarket led for most of the cycle. By August 2026, Kalshi held about 82% of combined volume across the three largest venues, and Bank of America analysts put its share of the regulated US market near 91%.
The World Cup inflated July into a record, and August was the first clean print afterward.
Source: The Block prediction-market dashboard, accessed 2 September 2026. July is the revised series, which runs about $2.4B above the snapshot published on 3 August.
What that volume actually is matters more than its size. Kalshi told a federal court that sports made up 80% to 90% of its listed contracts and its revenue. Polymarket spreads wider, at roughly 39% sports, 32% politics and 20% crypto since July 2024, per Pew's read of The Block data. A venue that is four-fifths sports operates as a betting exchange with continuous events and small repeat tickets, which is precisely the product profile the CFTC's own proposed rule flags as hardest to defend on public-interest grounds.
The volume numbers deserve less trust than the valuations imply
Every headline in this sector rests on a number nobody audits. Sethi's wash-trade detection, which combines volume-to-capital ratios with counterparty network structure, found stretches where up to 60% of Polymarket volume was traders transacting across wallets they controlled. The rate climbed around the anticipated token distribution and falls where fees apply.
His verdict to us was measured and damaging:
"Polymarket volume claims should be treated with a bit of skepticism."
Definitional chaos compounds it. Robinhood's own Q2 release reported 13.6 billion event contracts through its hub and more than 3.5 billion on Rothera, then warned that the two figures cannot be added because each counts a different thing. One company, one quarter, two incompatible definitions of the same word. The Block quietly revised a single month by roughly $2B while nobody was disputing anything. Anyone quoting a precise market size, platforms included, is reporting a definition as though it were a fact.
The forecasting case is real, narrow, and has little to do with crowds
The strongest evidence for prediction markets in 2026 came from a central bank. The Federal Reserve working paper on Kalshi's macroeconomic markets found well-calibrated, continuously updating density forecasts, a perfect record on the day before FOMC rate decisions, and errors on core CPI and unemployment almost identical to the Bloomberg consensus. That is a genuine result from a hostile witness, and it describes the narrowest possible case: liquid, frequent, cleanly observable events with deep professional participation.
Outside that band, the crowd story collapses. Roughly 3% of traders drive most price discovery. The top 1% of profitable accounts take 76.5% of profits. Whale, high-frequency and power-trader tiers hold 81.4% of notional across 12.6% of addresses.
Dr. Robin Hanson, whose scoring-rule work founded the field, told us the concentration finding describes markets generally.
"A lot of what many people don't like about prediction markets is what they don't like about ordinary financial markets, but maybe just don't know."
Sethi puts the mechanism's real edge somewhere the industry rarely advertises: events with no precedent, where statistical models have nothing to calibrate against. When Biden's June 2024 debate performance collapsed his contract price, the major forecasting models had to be suspended entirely, generating what Sethi calls zombie forecasts of a race that was no longer going to happen. The market repriced within the half hour.
"It's a mechanism that gives you something when you're in uncharted waters."
The integrity record is the year's actual story
The transparency pitch met its own year. These are documented cases with charges, dollar figures or federal probes attached.
Two entries deserve emphasis. Kalshi's surveillance caught the teleprompter trades and referred them, which is the regulated model working as advertised. And a Polymarket contract on an Iranian missile strike named a working Times of Israel journalist in its resolution rules; traders on the losing side sent him death threats demanding he rewrite the story. A market on a real event had become a financial reason to interfere with the reporting of it.
The oracle layer is the structural version of the same problem. Proposing a resolution on UMA required a bond of about $750 against a Zelensky market carrying up to $242M of exposure, adjudicated by a token whose market cap sat near $95M. The security budget was smaller than the thing it secured.
Washington says financial instrument. Almost everyone else says gambling.
The federal posture is the friendliest the sector has known. CFTC Chairman Michael Selig withdrew the Biden-era ban proposal and published a 267-page rule on public-interest determinations in June. The comment record that closed on 27 July is where it gets interesting: 44 state attorneys general filed jointly against it, and the NFL filed separately the same day.
Then the courts moved. A Third Circuit panel sided with Kalshi against New Jersey in April. District judges in New York and Connecticut went the other way, finding sports event contracts function as wagers and that federal law does not preempt state gambling statutes. New York sued for more than $36B. On 11 August the CFTC invoked emergency authority to keep Kalshi operating there. A circuit split is forming and Supreme Court review is live. Congress is circling from several directions, including a bill to bar lawmakers and their families from trading policy outcomes.
Outside the US the question is mostly settled against the platforms, with more than a dozen jurisdictions imposing blocks or blacklists, including France's ANJ ordering ISPs to block Polymarket. No jurisdiction anywhere has built a bespoke prediction-market regime, so every operator lives in the gap between gambling law and financial-instrument law.
Where we think the durable value sits
Our thesis is that the money is in the plumbing rather than the casino. Regulated clearing and surveillance, oracle integrity, institutional data and the structured-products layer are where demand survives a bad ruling. ICE bought a data feed, not a trading take, and that is the tell.
Distribution turned hostile in 2026, which is the year's most underrated lesson. Robinhood went from supplying nearly half of Kalshi's volume to about 17.5% by standing up its own exchange with Susquehanna. IG Group agreed to pay up to $1.3B for Underdog rather than keep renting access. Rented distribution behaves like a moat until the tenant buys the building.
A bet on more retail sports and politics velocity is a bet on the segment regulators are most likely to clip and the one the forecasting evidence supports least.
None of this is investment advice.
Dr. Robin Hanson takes up why organisations reject prediction markets in our futarchy episode (watch from 16:49, “Why organisations reject prediction markets”).