Kalshi Published a User Study. The Response Rate Was 13.8%. The Timestamp Was 2026.

Two numbers anchor the report Kalshi Research published under the title "Who Trades Prediction Markets?" The first is 325 — the count of complete responses pulled from a target pool of 2,360 funded accounts. The second is 2.42, the mean Berlin Numeracy Test score attributed to those respondents against a general-population baseline of 0.62. Read together, the report argues that prediction-market participants are self-taught, non-professional, and statistically sharper than the public. I have spent sixteen years reading platform-authored research, and my first instinct is not to celebrate the finding but to interrogate the sample. A 13.8% completion rate is not a dataset. It is a residue, and residues have a shape. Kalshi is not a blockchain protocol. That distinction matters before any of these numbers are reused. The platform operates as a designated contract market registered with the CFTC, settles in U.S. dollars through a centralized order book, and issues no token. Its structural opposite is Polymarket, which runs on Polygon, settles in USDC, and derives its transparency from on-chain state rather than from a regulator's stamp. The two represent competing trust models: one asks users to trust a licensed intermediary, the other asks them to verify a contract. When a centralized venue publishes a study of its own users, the audit trail stops at the press release. There is no block explorer for methodology. The timing is not incidental. Prediction markets entered an accelerated phase through the 2024 U.S. election cycle, and every operator in the category is now competing for the same scarce resource: legitimacy. For Kalshi, legitimacy is the product. The CFTC license is the moat, and the moat is only as deep as the regulator's willingness to defend it. A report demonstrating that users are rational analysts rather than gamblers is not neutral research. It is positioning, aimed squarely at the people who decide whether event contracts are derivatives or dice. The category's history is a study in regulatory whiplash. PredictIt operated under a CFTC no-action letter that was later withdrawn, forcing it to unwind its U.S. market. Polymarket paid a penalty and restructured U.S. access. Kalshi's path was different: it litigated the right to list election contracts and won, converting a regulatory question into a legal precedent. That victory is why a user study matters more than it otherwise would. A platform that has already litigated its own legitimacy now needs to defend it in the court of institutional opinion, where the currency is perception rather than statute. Now the forensics. Start with the response rate. 325 of 2,360 is 13.8%, and non-response bias is not a footnote — it is the dominant variable. The users most likely to complete a long survey are the users most engaged with the platform. That is a selection filter, not a random draw, and it systematically inflates every flattering metric in the document. I saw the same pattern in 2021 when I traced coordinated wash trading across fifteen wallets in the BAYC floor market. The visible transactions told a story of organic demand; the wallet clustering told a different one. Data doesn't lie, but datasets are curated. A platform that selects its own respondents is curating its own mirror. Then the sports exclusion. The report deliberately omits sports and "special" categories from the sample frame. This is the single most consequential methodological choice in the document, and it is buried. Sports is the highest-volume category in prediction markets by a wide margin. Remove it, and what remains is a population skewed toward analytical, event-driven, non-recreational contracts. The respondents were never going to resemble sportsbook users, because sportsbook users were filtered out before the first question was asked. The "rational trader" finding is partly an artifact of the filter, not an observation about the market. The income claim deserves the same scrutiny. The report aligns participant income with the median household income of their ZIP code, using American Community Survey data. This is a clever proxy and an ecological fallacy at the same time. ZIP code is not a person. A median income of $70,000 in a postal area says nothing about the individual trader who answered the survey, and the aggregation conceals the variance that would actually test the claim. The same logic applies to the professional-background numbers: 82% reporting no full-time traditional finance experience and 91% not trading full-time are only meaningful if the sample is representative, and the response rate tells us it is not. Consider also what is self-reported. The Berlin Numeracy advantage and the resistance to gambler's fallacy are both measured through the respondents' own answers. Self-reported competence is the least reliable instrument in behavioral research. Based on my audit experience, I treat self-reported data the way I treat unaudited code — as a hypothesis, not a finding. A real test would cross-reference survey answers against actual position sizing, realized P&L distribution, and loss-chasing behavior. None of that appears here. The 2.42 is a claim about a test, administered online, to volunteers, and compared against a telephone sample collected under entirely different conditions. The gap may be real. It may also be an artifact of administration. There is a structural point that cuts the other way, and it deserves stating. Kalshi issues no token, and that is a genuine advantage. On-chain prediction markets have historically been distorted by airdrop expectations — users farming volume they would never trade otherwise. Kalshi has no such incentive, because there is no token to farm. The participants in this study are trading dollars at risk, not points. That is a cleaner signal than any incentivized on-chain metric, and it is the one place where the centralized model produces better data than the decentralized alternative. The trade-off is that Kalshi also forfeits the token as a growth lever. Its user base has to expand on product and brand, not on speculation. Run the numbers against the framing one more time. The report's headline claim is that participants are "not who you think." That framing only works if the baseline expectation is a wealthy professional gambler. But the actual competitive baseline for Kalshi is Polymarket, whose volume leadership in political markets is well documented and whose users skew crypto-native and younger. Against that competitor, the interesting question is not whether Kalshi's users are rich or self-taught. It is whether a dollar-settled, KYC-gated, sports-excluded platform can scale a user base large enough to matter. The report answers none of that. It measures identity, not liquidity. The last structural caveat is the publisher. Kalshi Research is Kalshi. A platform studying its own users, publishing its own conclusions, with no third-party replication, has an interest in the result. That does not make the data false. It makes the data interested, and interested data requires independent verification before it becomes evidence. The detail that stops me cold is the timestamp. The report is marked for October 2026, with the survey distributed in September 2026 and administrative data covering activity since January 2026. If today is earlier than those dates, the document contains future dates — which points to a template placeholder, a mislabeled source, or synthetic content. That is not a small thing. It undermines the credibility base of every figure downstream. Verify the hash, ignore the hype. Before a single number from this report is cited, the publication date needs to be resolved, because an unverified timestamp invalidates the audit trail the same way an unverified block invalidates a settlement. I have held this line since 2017, when I spent six weeks auditing the Ethereum Classic block-reward logic after the 51% attack. The lesson from that report was simple: a finding without a verifiable chain of custody is a rumor with a spreadsheet. There is also a causal trap the report walks into without acknowledging it. The narrative implies that prediction markets attract people who are better at probabilistic reasoning. The alternative is equally plausible: that participation trains that reasoning, or that a numeracy test administered online to self-selected volunteers simply outperforms a random telephone sample for reasons that have nothing to do with trading skill. Correlation is being dressed as capability, and the dressing serves a marketing purpose. I made this distinction once in the other direction. In 2020 I flagged abnormal gas-fee spikes preceding exploits and called the Mango Markets collapse three days out. The signal was real, but I was careful to label it a pattern, not a law. The report is less careful. It reads a selection effect as a capability proof. On-chain metrics > Twitter polls, and a self-published survey sits somewhere between the two. The useful signal here is not that prediction-market users are smart. It is that a regulated venue is now willing to spend research budget arguing that its users are not gamblers — a tell about where the regulatory pressure is concentrated. Watch the CFTC's posture on event contracts, not the numeracy scores. If the compliance moat holds, this study becomes a template for how centralized venues market themselves against on-chain competitors. If it breaks, the 2.42 will be the least interesting number in the file.