Most of the argument about prediction markets in the United States is an argument about who decides. The amended preliminary injunction entered against KalshiEX on 12 August 2026 in King County Superior Court contains that argument, and we have already written about how the order sorted the order book by the platform's own category tabs. But the order also contains four paragraphs that have nothing to do with preemption, swaps or the Commodity Exchange Act. Findings of fact 18 to 21 of the order published by the Washington Attorney General are a public health record, and they are the part of the document that no appellate ruling about jurisdiction will touch. They are also the part the category has no answer to, because answering them requires numbers that no venue publishes.
Four findings that do not mention jurisdiction
Finding 18 quotes the Washington Legislature's own words in RCW 9.46.071, that "some individuals in this state are negatively impacted by problem gambling and gambling disorder", and sets beside them the view of the Washington Supreme Court in Rousso v. State, 170 Wn.2d 70 (2010), where the court called internet gambling "immediately accessible at home" and treated that accessibility as a reason the Legislature could ban it outright. Finding 19 supplies the arithmetic. Finding 20 cites a review by Nerilee Hing and colleagues on gambling and domestic and family violence in Frontiers in Psychology, October 2022, for the proposition that the harm reaches people who never traded. Finding 21 cites academic work by Nizan Geslevich Packin and Sharon Rabinovitz for the claim that prediction markets share "structural and functional similarities" with online gambling while carrying "social and commercial legitimacy".
One note on that last citation, because the order makes it hard to find. The order names the journal as Nature. The Packin and Rabinovitz piece, Prediction markets as a public health threat, was published in Science on 16 April 2026, in the volume and issue the order gives and under the DOI the order links. A journal name slip, not a substantive one, but worth knowing if you go looking.
The percentages the order quotes, and where they come from
Finding 19 is built on the 2021 Washington State Adult Problem Gambling Prevalence Study, written by Rachel A. Volberg of Gemini Research for the Washington State Health Care Authority. The numbers the order quotes are in the report: among Washington adults who gambled in the past year, 3.5 percent scored in the moderate to severe range on the Problem Gambling Severity Index, 1.5 percent of all adults; and the rate among self-reported online gamblers was 10.3 percent against 2.6 percent for people who gambled only at land-based venues. That is the comparison the order rests on, and it is quoted accurately.
The order's pin cite is to pages iv to vii, the executive summary. The qualifications sit further in. Table 10 of the report puts 24 unweighted respondents behind the 10.3 percent figure, with a 95 percent confidence interval running from 5.3 to 15.3 percent. The 2.6 percent comparison is derived rather than observed: the report states that "if we assume that all of the gamblers who did not say Yes to the online gambling question were brick-and-mortar gamblers, the problem gambling prevalence rate for brick-and-mortar gamblers was 2.6%". The same paragraph adds that "further analysis is needed to test this difference statistically due to small group sizes". The limitations section notes a 19.2 percent response rate, warns that subgroup estimates "should be viewed with caution since they may be unreliable", and says a cross-sectional survey "limits our ability to draw any cause-and-effect conclusions".
Why a methodology complaint is the wrong response
It would be easy to turn the preceding paragraph into a brief. It would also be a mistake, and it is worth being clear about why. A finding of fact in a preliminary injunction does not need to survive peer review; it needs to be the best evidence in front of the judge. On the question of whether an always-open, phone-sized venue carries more risk than a building you have to drive to, the state brought an address-sampled survey of 9,249 respondents using the dominant international instrument. The venue brought nothing comparable, because nothing comparable exists. In that matchup, a confidence interval is not a defence. The only durable answer to a number is another number.
The author of the Washington report saw this coming, in a sentence about her own field: when prevalence results are published, "policy makers and the media generally focus their attention on a single number".
What the 2021 survey could not have been looking at
The survey was fielded between July 2021 and early September 2021. Online gambling was illegal in Washington throughout, which is why the report flags, as one of two notable findings, that 11.5 percent of gamblers reported gambling online anyway. The people in the 10.3 percent group were therefore, by construction, using venues with no state-supervised deposit controls, no state self-exclusion hook and no supervised settlement. Regulated event contracts were not in that picture: the Commission's own account in the June 2026 proposal on event contracts, at 91 FR 35846, dates the substantial expansion in event contract submissions and trading activity to October 2025 and puts notional volume at 25 billion dollars in March 2026.
That cuts in one direction only if you finish the sentence. The number describes a channel rather than this product. But a venue cannot say so credibly while publishing nothing about its own channel, because "those people were somewhere else" is a claim about data, and the party holding the data is the venue.
A single yes or no question did the sorting
There is a design detail in the Washington report that matters more for prediction markets than the headline percentages. Respondents were classified as online gamblers by one question, number 21, "In the past 12 months, have you gambled online?". The report then tested that self-label against a later section listing specific online activities, and the two did not line up. Of self-described online gamblers, 73 percent were confirmed by their own activity answers, 12 percent selected none of the listed online activities, 6.7 percent selected only gaming and 8.6 percent only playing games on a device. Going the other way, 22.4 percent of respondents who answered No to question 21 still selected activities the report classes as gambling online. The report draws the obvious conclusion, that adults "were not always aware that these activities have an element of gambling".
A survey has to ask people what they did. An exchange does not. Every fact a prevalence study reaches for through self-report, time on the product, deposit frequency, escalation, a repeat session after a loss, sits in a venue's own records with a timestamp on it.
Seven families of indicators, published and in force since 2023
Great Britain's Gambling Commission, in its formal guidance under Social Responsibility Code Provision 3.4.3, in effect since 31 October 2023, requires remote licensees to use indicators that "must include" seven categories: customer spend, patterns of spend, time spent gambling, gambling behaviour indicators, customer-led contact, use of gambling management tools, and account indicators. The guidance calls this "the minimum requirement and therefore not exhaustive". Its worked examples are specific enough to code against: binge deposits, deposits clustered on payday, overnight play, lengthening sessions, chasing losses, limits "set so high as to be meaningless", failed deposits, multiple payment methods, and a refusal to use management tools.
None of that requires a survey, a diagnosis or a clinician. It is arithmetic over account and order records that a designated contract market already keeps. And the definitional work a venue would otherwise have to invent, then defend as self-serving, has been done in public by a regulator with twenty years of returns behind it. Adopting someone else's published definition is the cheapest credibility available.
The Commission asked for exactly this, once, in one paragraph
The federal regulator has already named the measurement. In the public interest discussion of the June 2026 proposal, at 91 FR 35857, the Commission lists trading characteristics "generally associated with addictive potential", among them outcomes at unpredictable intervals, perceived skill, near miss experiences and loss chasing, adds low minimum position sizes, continuous availability and engagement notifications, and then writes that "prediction markets could monitor calibration and cohort outcomes and document when remediation, such as product redesign or targeted education, reduces these adverse consequences".
That sentence is the brief. It is also almost the whole of it. Across the proposal, which runs to roughly 76,500 words by our own extraction of the Government Publishing Office text, the word stem "addict" appears once, in that paragraph. "Problem gambling" appears zero times, "prevalence" zero, "deposit limit" zero, and "self-exclusion" once, in the same passage. On the reporting side the picture is the same: 17 CFR 16.01 requires exchanges to publish volume, gross open interest and prices, and says nothing about people, which is the gap we described when three million accounts arrived and nobody counted who stayed. Waiting for a rule to compel harm measurement means waiting for a rule that, as of 8 October 2026, exists as one conditional sentence in a cost-benefit section.
Publishing play data has been done, and it did not end in a confession
The objection to publishing is that the numbers will be used against you. There is a precedent worth knowing. In a study by Broda, LaPlante, Nelson, LaBrie, Bosworth and Shaffer, published in Harm Reduction Journal in 2008, the betting operator bwin gave researchers two years of actual account records for 47,000 subscribers to test whether its deposit limits changed behaviour. The headline result was about the control, not the customers: only 160 subscribers, 0.3 percent, ever exceeded a deposit limit at all.
Read against Directive 03, which commits signatories to deposit limits, cool-off periods and self-exclusion as product features rather than concessions, that finding is the useful kind of uncomfortable. A limit that almost nobody reaches is not evidence of a safe population; it is evidence that the limit is set where it binds on nobody. You cannot learn that from a prevalence study, and you cannot learn it without publishing.
The cross-venue register already exists and reports no total
The infrastructure question is further along than the measurement question. Kalshi's help centre documents three user-side controls: a trading break, a personalised funding cap on monthly deposits that "cannot be changed or lifted until the limit has reached its expiration date", and voluntary self-exclusion, which the same page says also routes into a broader scheme. That scheme, SelfExclude.io, operated by IC360, states that Kalshi and Novig are fully integrated and that integrations with Polymarket, Robinhood, ProphetX, Matchbook and Juice Exchange are in progress, offers terms of one, three, six or twelve months, and describes matching by double-hashed identifiers so that "platforms query our system using the same hashing method to check for matches" without receiving names or ID details.
Two things follow, as of 8 October 2026. The privacy argument against measurement is already answered by the category's own engineering: a shared register that works on hashes proves that aggregate reporting, which needs no identifiers at all, is the easy case. And the register publishes no count. Not enrolments, not duration mix, not how many matches it blocks. Kalshi's pages also note that neither a trading break nor a self-exclusion limits access through a futures commission merchant, and that it is "ultimately the trader's responsibility not to trade during this time", which is a measurable gap rather than a rhetorical one.
Washington's own list stops at the card room
There is a symmetry in the order worth sitting with. The statute the court cites in finding 18 is the same statute that builds Washington's measurable remedy: RCW 9.46.071(2) directs the state Gambling Commission to run a statewide self-exclusion programme on a single form, with a route for federally recognised tribes to opt in. Its statutory minimum covers establishments offering house-banked social card games, and the Gambling Commission's own page for the programme describes it as excluding you "from gambling in all card rooms statewide".
So the state's one enforceable, countable instrument does not reach the channel its own finding singles out, while a voluntary scheme in this category already spans two venues with five more in progress. That is not a gotcha at Washington's expense; the state banned the channel instead. It does mean the category is holding a better instrument than the jurisdiction judging it, and is saying nothing about what the instrument shows.
Six numbers a venue could put on a page this quarter
Everything below is aggregate, needs no new rulemaking, no identifiers and no clinical claim, and maps onto definitions someone else has already published and defended.
- Tool uptake. Share of accounts that traded in the quarter with an active funding cap, trading break or self-exclusion, split by the three.
- Whether the cap binds. Share of capped accounts that reached their cap, which is the bwin question asked of this product.
- Indicator counts. Accounts crossing each of the seven British indicator families, with the thresholds stated, and what action followed each crossing.
- Revenue concentration. Share of net revenue from the top one percent of accounts ranked by twelve-month net loss, which is the single number that tests Directive 06.
- Calibration and cohort outcomes. Realised frequency against price bucket, and net outcomes by joining cohort, which is what the Commission asked for at 91 FR 35857.
- Register coverage. Self-exclusion requests received and honoured, and how many were reachable through intermediated access rather than only the venue's own app.
None of these six is a concession. Four are operating metrics any venue already computes internally to run a risk desk, and the other two are published definitions waiting to be applied. We have argued before that the real fight over event contracts is a fight about who sees what. This is the same fight at the level of the user.
A finding of fact is answered with a measurement
Washington's order is in effect, and the broader dispute over whether states may reach these contracts at all is still moving through the federal courts. Whichever way that goes, findings 18 to 21 will be reused, because they are cheap to reuse: a published state survey, a thirty-year-old instrument and a court record are now sitting there for the next attorney general who wants them. The jurisdictional question will be settled by someone in a robe. The health question will be settled by whoever shows up with data, and at the moment only one side is holding any. A venue that can say, with stated thresholds and a stated period, how many of its accounts crossed each indicator family and what it did about them has moved the argument from whose survey to whose record.