This is the third of three posts this week on prediction markets. Monday was about information — the signal that travels before the message. Wednesday was about jurisdiction — the classification question that reproduces itself at every level you try to answer it. Today is about something prior to both: whether the classification of a prediction market contract as a futures contract, rather than a wager, is a finding about what the contract is, or a decision about what it becomes.
On Monday, the Third Circuit ruled 2-1 that New Jersey cannot enforce its gambling laws against Kalshi’s sports event contracts — the first federal appellate decision on the national question. The majority held that Kalshi’s DCM license and the Commodity Exchange Act’s preemption provisions likely shield the company from state enforcement. The dissent, by Circuit Judge Jane Roth, called the majority’s reasoning an act of “legal alchemy” — a classification that transforms what is, in practice, sports gambling into a federally protected financial instrument by the application of a label. Judge Roth’s dissent is, in the formal sense I want to develop here, the more interesting opinion. Not because it is right on the law — that question is genuinely hard and will reach the Supreme Court — but because it names the underlying problem precisely.
The descriptive and the constitutive
There is a distinction in philosophy of language between descriptions and constitutive declarations. A description reports a state of the world: “it is raining.” A constitutive declaration brings a state of the world into existence: “I now pronounce you married.” The difference is not merely grammatical. A description can be true or false — it is true if and only if it accurately represents what is the case. A constitutive declaration does not describe an antecedently existing state; it creates one. Whether it succeeds depends not on its correspondence to facts but on whether it was made by the right person, in the right context, according to the right procedure.
Most classifications are descriptive. A physician who classifies a tumor as malignant is making a claim about what the tumor already is. A meteorologist who classifies a storm as a Category 3 hurricane is applying a pre-existing scale to a pre-existing storm. The classification can be right or wrong. If it is wrong, the correction is to reclassify, not to pretend the tumor was never malignant or the storm was never a Category 3.
Some classifications are constitutive. A judge who classifies a contract as legally enforceable is not reporting a property the contract possessed before the ruling. The enforceability is created by the ruling. A legislature that classifies an activity as a federal felony is not describing a pre-existing moral category; it is creating a legal one. The CFTC, when it classified Kalshi as a Designated Contract Market, was not reporting that Kalshi’s contracts were already swaps in some mind-independent sense. It was establishing that, for purposes of federal law, they would be treated as swaps — which meant they would be subject to the CEA, not to state gambling law.
Judge Roth’s “legal alchemy” formulation is precisely this complaint: Kalshi obtained a constitutive classification from a sympathetic regulator and is now presenting it as though it were a descriptive finding about the nature of the contracts. The CFTC said “swap.” Therefore — the argument runs — the contracts are swaps, in the sense that makes state gambling law inapplicable. But the CFTC’s classification did not discover the contracts’ essential nature. It conferred a legal status. And the question of whether that conferral validly preempts state law is not answered by pointing to the conferral.
Why this matters beyond Kalshi
The distinction between descriptive and constitutive classification is not an exotic philosophical point. It is a structural feature of how regulatory systems work, and the failure to maintain the distinction produces a specific and recurring pathology: classification capture.
Classification capture occurs when the entity being classified obtains the power to influence its own classification, and uses that power to obtain a constitutive classification that confers legal advantages the entity could not otherwise claim. The classification looks descriptive — it has a label, a regulatory category, a statutory hook — but it is functionally constitutive, because the entity’s behavior did not conform to the category; the category was extended to cover the behavior.
The Nevada state court that ruled April 3 saw this clearly. Judge Woodbury’s finding that Kalshi’s contracts are “indistinguishable” from sports wagers was a descriptive classification: he looked at the activity, applied the pre-existing legal definition of wagering, and reported what he found. Kalshi’s response — that the CFTC’s DCM license settles the question — is an appeal to a constitutive classification that was issued before the question was put in its current form. The DCM license says Kalshi is a designated contract market. It does not say, and arguably could not say, that every contract Kalshi lists is therefore a swap beyond state regulatory reach. The Tennessee court that ruled for Kalshi in February made the constitutive move: it treated the license as settling the classification. Nevada declined to make that move and looked at the thing itself.
This is not merely a disagreement about law. It is a disagreement about the metaphysics of classification — about whether the legal category precedes and determines the regulatory treatment, or whether the regulatory treatment should follow from an independent assessment of what the thing actually is.1
The menu and the message
There is a formal result that connects this to the signaling literature from Monday’s post, and it is worth making explicit. Before stating it, one observation worth flagging for later: every non-trivial classification system has a logical flaw embedded in it somewhere — inputs that fall exactly at the edge of the category’s definition and cannot be cleanly processed by the system’s own rules. In the context of classifiers specifically, these are sometimes called “edge cases” or “boundary cases,” and they are not bugs to be patched — they are structural features that reveal what the system actually is. We will come back to this.4 In signaling models, the equilibrium interpretation of a signal depends on the menu of available signals — on what the sender could have sent but didn’t, as well as what they did send. A sender who can costlessly obtain any label they want provides no information through their choice of label, because the label no longer separates types. The receiver cannot infer anything from the label because every type would choose the same label if they could.2
The CFTC’s current posture toward prediction markets — issuing DCM licenses to platforms, asserting exclusive federal jurisdiction, declining to distinguish between contracts that function as derivatives and contracts that function as wagers — collapses the separating equilibrium. If any platform can obtain a DCM license and thereby claim federal preemption of state gambling law, the license no longer signals anything about the platform’s actual activity. It signals only that the platform applied for and received the license. The informational content of the classification goes to zero.
This is what Judge Roth’s dissent means by “legal alchemy.” Not that the CFTC made a factual error. That the CFTC’s classification, applied in this context, transforms a signal that was supposed to carry information about the nature of financial instruments into a signal that carries information only about regulatory access. The label means “we are federally regulated” rather than “we are what the federal regulatory category was designed to cover.”
A prediction market that classifies itself as a financial exchange, and uses that classification to escape regulation as a gambling platform, has done something the signaling literature describes with precision: it has obtained a separating equilibrium outcome — the legal treatment reserved for financial exchanges — without paying the separating cost that was supposed to make the label credible.3
The three questions this week — what kind of signal was the Iran trade, who gets to answer the jurisdictional question, and whether calling something a swap makes it one — are the same question in three registers. All three are asking whether the label precedes and determines the thing, or whether the thing precedes and constrains what labels can legitimately be applied. Courts are going to be answering that question, in one form or another, for years. The Supreme Court will eventually answer it definitively. What it will not do — what no answer can do — is make the distinction between descriptive and constitutive classification go away. The problem is not legal. It is structural. This is the first of what I expect will be an occasional series on exactly this kind of problem — the cases that fall at the edges of classification systems and reveal their structure. I’m calling it “the Junk Drawer.” (Ed: That’s either the most honest or the least flattering name for a research program I’ve ever heard…wait, is there ever really a difference?)
With that, I leave you with this.
1 The philosophical literature here runs from J.L. Austin’s speech act theory through John Searle’s work on institutional facts and constitutive rules. The relevant distinction for legal purposes is between what Searle calls “brute facts” — facts that obtain independent of any human institution — and “institutional facts” — facts that obtain only within a system of constitutive rules. Kalshi’s contracts are brute facts: money changes hands contingent on outcomes. Whether they are swaps or wagers is an institutional fact, constituted by legal rules. The dispute about which institutional fact obtains is a dispute about which constitutive rules apply, which is exactly the jurisdictional question from Wednesday’s post. The recursion is not accidental. Constitutive classification questions are inherently jurisdictional.
2 This is the pooling equilibrium, in the taxonomy of signaling games. See Spence (1973), and the discussion in Monday’s post. A pooling equilibrium obtains when all types send the same signal, rendering the signal uninformative. The separating equilibrium — in which the signal genuinely differentiates types — requires that the signal be differentially costly across types. When the CFTC issues DCM licenses without discriminating between platforms whose contracts genuinely function as derivatives and platforms whose contracts function as wagers, it eliminates the differential cost and induces pooling.
3 There is an analogy here to the sports leagues’ longstanding resistance to prediction markets on officiating decisions — not merely because of corruption risk, but because pricing a referee’s call changes the incentive structure around that call in ways that cannot be cleanly separated from the call itself. The market doesn’t observe the official neutrally; it alters what officiating means by attaching financial stakes to it. This is a version of the endogenous base rates problem from An Agenda-Setter with a Ticker: the classifier shapes what it classifies. Classification is never purely descriptive when the classified party can observe the classification and respond to it.
4 The formal home of this observation is Can a Game Know Its Own Rules?, which develops the connection between Gödelian incompleteness and the structural limits of rule systems. The short version: any sufficiently rich classification system will contain statements — inputs, cases, contracts — that the system cannot classify from within its own rules without expanding the system, at which point the new system has its own boundary cases. The Kalshi contracts are not at the boundary by accident. They are at the boundary because the boundary is where the interesting problems live.
1 thought on “What You Call It Is What It Is”
Comments are closed.