In July 2026 Kalshi confirmed to CNBC that about three million people opened accounts on the exchange during the World Cup. The tournament ran from 11 June to 19 July 2026, and the same reporting put more than 1.2 billion dollars through the single market asking who would lift the trophy, a record for one market on that venue. Those are the company's own figures about the company's own business, and there is nothing wrong with being pleased about them.
This piece is about the question that comes immediately afterwards. As of 16 September 2026, how many of those three million are still trading? Nobody outside the exchange can answer that. No filing carries the number, no rule asks for it, and no venue in this category publishes it voluntarily. Monthly volume is public, search interest is public, and the cohort itself is invisible.
We are not raising this because growth is suspect. A tournament that brings millions of people into a regulated market is a good thing for a category that spent years arguing it deserved to exist. We are raising it because an arrival number without a survival number is the easiest kind of figure to celebrate and the hardest kind to learn anything from, and because the people who arrived at the loudest possible moment are precisely the ones the sixth directive of our manifest is about.
Three million is an arrival count, not a participation count
An account opened during a tournament tells you that a marketing campaign worked. It tells you nothing about whether the person traded twice, understood what they bought, or came back in September. Sportsbook Review reported the same figure on 21 July 2026 in the same form, as a number the operator confirmed to CNBC, which is worth noting because it shows how this kind of statistic travels: it is announced once, at the peak, by the party with the strongest interest in it being large, and then it is never revisited by anybody.
That is not a complaint about either company. It is a description of an information gap that the category has not yet decided to close. Compare it with any other measure a venue puts out. Volume gets a monthly figure. Open interest gets a daily one. The size and fate of a cohort, the single thing that would tell you whether a marketing spike produced participants or produced churn, gets one triumphant sentence in July and silence thereafter.
August measured the tide going out, and it measured the wrong thing
The reversal was visible almost immediately. The Block reported on 2 September 2026 that combined August volume across Kalshi, Polymarket and Polymarket US fell 14.5 per cent to 45.33 billion dollars, the first month on month decline in a year, with 37.17 billion at Kalshi and 8.16 billion across the two Polymarket entities. Attention fell further than money did. The Defiant tracked worldwide Google Trends interest in the term prediction market and found that the index sat at 17 in the week of 9 to 15 August 2026 against a peak of 100 in the week of 7 to 13 June, a fall of 83 per cent.
Both figures are real and both are beside the point we are making. Volume can fall because the same users trade less, because fewer users trade the same amount, or because one market maker changed its inventory policy. Search interest measures the curious, not the funded. We have written before about why a volume record and a falling open interest can describe the same month, and the same warning applies here with more force: none of these series is a headcount, and no combination of them becomes one.
So the category finds itself in an odd position. It can tell you, to two decimal places, how many billions changed hands in August. It cannot tell you whether the three million people who arrived in June are twenty thousand people or two million people today.
One chain, one researcher, and the only public picture of a cohort
There is exactly one body of public evidence about what happens to prediction market users over time, and it exists only because one venue settles on a public blockchain. On 6 April 2026 the independent on chain researcher Andrey Sergeenkov published an analysis of 2.5 million Polymarket trader addresses with data as of 1 April 2026, drawn from Polygon transactions through Dune Analytics. The Defiant covered it on the same day, reporting the headline finding that 84.1 per cent of those addresses were in the red.
The profitability numbers are the ones that get quoted. The retention numbers are the ones that matter for an event cohort. In that dataset, 73 per cent of traders were active for no more than two months. Among the roughly 6,600 addresses that averaged more than 5,000 dollars of monthly profit, only 2.6 per cent were still active after a year. Two per cent of all addresses had ever made more than 1,000 dollars, and 0.033 per cent, some 840 addresses, had made more than 100,000.
The analysis also contains the closest thing this category has to a precedent for the summer of 2026. Sergeenkov notes that user growth surged around the United States election cycle of November 2024 and that profitability fell across every threshold as it did, on the straightforward reasoning that less experienced arrivals trade less successfully than the people already there. An event brings in a cohort; the cohort trades worse than the incumbents; most of the cohort is gone within two months.
We are deliberately careful with this evidence. It covers one venue, it counts only realised profit and loss, and open positions in unresolved markets appear as losses even though the money is merely locked, a limitation the researcher states himself while noting that 96 per cent of the volume involved markets that had already settled. It is a single study of a single order book, not a census of the category. It is also, as of 16 September 2026, the only thing anybody has.
The rulebook counts contracts, and Britain counts accounts
None of this is anyone breaking a rule, and that is the part worth sitting with. Under 17 CFR 16.01 a reporting market records and reports total volume of trading, total gross open contracts, and opening, closing and settlement prices. The regulation is careful and detailed about contracts. It asks for nothing at all about the number of accounts, customers or traders behind them. A United States venue that never said a word about its user base would be in perfect compliance.
A different regulator, looking at a neighbouring product, made the opposite choice twenty years ago. The Gambling Commission requires its licensees to report, on their quarterly regulatory returns, the number of accounts that actually got used. The guidance is one sentence:
"Record the total number of accounts that have been used by customers to gamble on Commission licensed activities in the 3 month period covered by the regulatory return, for GB customers only. Do not include suspended or dormant accounts."That definition is available in full on the Commission's regulatory returns guidance, and the interesting part is the exclusion. Dormant accounts do not count. Britain decided that the meaningful unit was a person who traded in the period, not a person who once signed up.
We are not proposing that American exchanges be placed under British gambling licensing, and readers of this site know how firmly we think the two perimeters should be kept apart. The point is narrower and harder to argue with: the measurement problem has a solved definition sitting in public, written by a regulator that has been counting this exact thing for two decades, and nothing prevents a venue from adopting it voluntarily tomorrow.
What the tournament crowd met on the way in
Directive 04 of our manifest says that a first time user "learns what a price means, how a spread costs them money and how their market resolves before they learn how to deposit faster." A tournament is the hardest possible test of that sentence, because the arriving user is in a hurry, the marketing is loudest, and the match starts in forty minutes.
So we checked one venue's public explanation on 16 September 2026, months after the crowd went through it. Kalshi's help centre is organised into named collections, and the front page lists eight articles under Getting Started against eighteen under Deposits and Withdrawals. The two Getting Started articles that explain the product do explain it: What are prediction markets? sets out that a contract pays one dollar if you are right, and How are prices determined? explains the price as an implied probability, with seventy cents for a market that thinks an event is seventy per cent likely. Neither of them mentions the spread, the fee, or what it costs to cross from one side of the book to the other. The venue does publish that: there is a clear article on fees, and it lives in the Trading collection, one level past the place a new account holder starts.
This is a sequencing observation, not an accusation, and we want to be precise about the difference. The material exists, it is written plainly, and it is free to read. Directive 04 is not a demand that the information be published somewhere. It is a claim about order: the cost of trading should reach the user before the deposit screen does, because after the deposit screen the user is no longer reading. A help centre is also not the same thing as an onboarding flow, and we could not observe what the June cohort actually saw inside the app during the tournament. We can only observe where the explanation sits today, and invite anyone who runs one of these surfaces to check their own order of operations against that sentence.
The surrounding layer matters at least as much, and we have written about the affiliate and partner screens that many users meet before they ever reach a venue. A venue can get its own sequence right and still hand the first explanation of its product to somebody paid per signup.
The incentive deadline that fell on 14 September
Two days before this article, a date passed that nobody marked. On 12 August 2026 the Division of Market Oversight issued an advisory on the self certification of incentive programs, published in full as Staff Letter 26-23. Most of the coverage at the time, including our own, concentrated on what the letter says about volume based rewards and wash trading. Footnote 15 contains a deadline, and it is worth quoting because it appears nowhere in the press release:
"Amendments to such programs that are reasonably within the scope of the advisory should be submitted via Rule 40.6(a) or Rule 40.6(d), for notice of non-substantive revisions, by September 14, 2026."
The same letter names, among the structures the division considers problematic, the practice of "selectively offering retention bonuses" alongside threshold bonuses, unlimited rebates, risk free trades and randomised rewards. Read that list next to a three million person cohort acquired during a single tournament and the connection is not subtle. Retention bonuses exist because somebody has already worked out that the cohort leaves.
The recommendation is staff guidance rather than a rule, and it directs filings to the Commission's portal rather than to a public announcement, so the absence of news is not evidence that exchanges ignored it. We checked the Commission's press releases on 16 September 2026 and found nothing published about the date in either direction. What a venue chose to amend by 14 September is, for now, another thing the public cannot see.
What belongs on a quarterly cohort page
None of what follows requires a rulemaking, a court to rule, or a competitor to move first. All four are aggregate, none exposes a trading strategy, and any venue could put them on a quarterly page.
- Cohort size by month of first trade. Not accounts opened, accounts that traded, using the Gambling Commission's exclusion of dormant accounts or a stricter one of the venue's own.
- Survival at 30, 90 and 180 days. What share of each monthly cohort placed a trade in the later window. A single line per cohort makes an event spike and its aftermath visible without revealing anything about any individual.
- Distribution of realised outcomes by cohort. The share of each cohort in profit and in loss at the end of the period, which is the figure that on chain researchers already compute for one venue and nobody computes for the rest.
- Time from first deposit to first trade. A blunt but honest proxy for whether education preceded the transaction, and the only one of the four that speaks directly to Directive 04.
A venue that published these during a quiet month would be handing its competitors nothing and its regulators a great deal of goodwill. A venue that published them for the June 2026 cohort specifically would be the first firm in this category to tell the public what a tournament actually produced.
Arrivals are a milestone, the cohort is the test
There will be another one. There is an NFL season under way, a Super Bowl in February, and a Winter Olympics after that, and each will bring its own spike, its own advertising, its own bonus codes and its own triumphant arrival figure. The category will have the same conversation about volume records in the month after, and unless something changes it will have nothing to say about the people.
Our position has not moved. We want these markets to exist, we want them to grow, and we would rather they grew through a tournament than not at all. But a standard that only measures money is a standard that will be written by somebody else the first time a cohort goes badly wrong in public. The number that tells you whether growth was real is already defined, already collected by a regulator next door, and already computable by any venue that wants to. Publishing it before someone demands it is the entire difference between setting a floor and being handed one.