This is the second of three posts this week on prediction markets. Monday’s post was about information — specifically, about what kind of signal moves through a prediction market before an official announcement arrives. Today’s is about jurisdiction. Friday’s will be about something more fundamental than either.
On April 3, a federal judge in Arizona was supposed to rule on whether Kalshi’s classification as a Designated Contract Market under the Commodity Futures Trading Commission preempts Arizona’s authority to prosecute the company for operating an illegal gambling business. He didn’t. Instead, he raised a prior question: whether the federal court has the authority to answer that question at all, given that state criminal proceedings against Kalshi are already underway. The Anti-Injunction Act, he noted, generally bars federal courts from enjoining ongoing state court proceedings. The question of whether any of the three statutory exceptions apply is now before him.
This is not a procedural detour. It is the argument, stated in legal form. The question of who gets to classify Kalshi’s activity does not have a stable answer at any single level of the jurisdictional hierarchy — and the attempt to find one by going up a level simply reproduces the same question at the new level. It is, in the formal sense that this blog has used before, a conservation problem. The classification question doesn’t get resolved. It relocates.
The classification that does all the work
Start with the load-bearing classification. Kalshi holds a Designated Contract Market license from the CFTC — the same category of license held by the Chicago Mercantile Exchange. Under the Commodity Exchange Act, DCMs are subject to exclusive federal jurisdiction. That classification is not incidental to Kalshi’s legal position. It is Kalshi’s legal position. Every argument the company makes in every state lawsuit begins and ends with it: we are a federally regulated exchange, not a bookmaker, and the distinction is not ours to make or yours to contest — it was made by the federal government when it issued the license.
Arizona’s response is that the classification doesn’t settle the underlying question. A DCM license authorizes Kalshi to operate futures markets. Whether the contracts Kalshi is actually selling — on election outcomes, on whether Iran will strike a military target, on whether a specific athlete will be injured — are genuinely futures contracts, or are instead wagers dressed in financial language, is precisely what is in dispute. The CFTC classified the instrument. Arizona is contesting whether the instrument is what the CFTC said it was. The license is the conclusion of a prior classification, not an answer to the question being asked.
This is the first level of the recursion.
All the way down
The structure of the jurisdictional dispute, set out explicitly, looks like this.
At the first level: is a Kalshi event contract a futures contract or a wager? The CFTC says futures contract. Arizona says wager. Nevada’s state court, ruling April 3, said wager. Tennessee’s federal court, ruling in February, said futures contract. The question has no settled answer and the courts addressing it are not bound by each other’s conclusions.
At the second level: assuming the first-level question were answered — assuming someone authoritatively determined that these contracts are futures, or wagers, or something else — which court or agency has the authority to make that determination in a way that binds others? Kalshi says the CFTC already made it, via the DCM license. Arizona says state courts have independent authority to classify activity under state law regardless of federal licensing decisions. The CFTC, in its amicus brief to the Ninth Circuit, says its jurisdiction is exclusive. Massachusetts said in January that “overly broad” is a precise description of that theory.
At the third level: the Anti-Injunction Act question the Arizona federal court raised on April 3. Even if the federal court would otherwise have authority to address the second-level question — even if federal preemption analysis is squarely within federal jurisdiction — can the federal court exercise that authority when the state has already initiated criminal proceedings? The Anti-Injunction Act says federal courts generally cannot enjoin state court proceedings. Kalshi’s workaround — shifting from “enjoin the state” to “preclusion,” arguing that a federal ruling on the merits would bind the state court without technically enjoining it — was described by Arizona’s lawyers as “a stunning concession” that the federal lawsuit is circular: Kalshi is asking the federal court to issue a ruling that would control the outcome of proceedings it admits it cannot directly stop.1
At the fourth level: who determines whether the Anti-Injunction Act applies, and whether Kalshi’s preclusion theory is valid? The Arizona federal court — whose own authority to reach the merits is precisely what is in question. The court is being asked to classify its own jurisdiction over the jurisdictional question.
Turtles all the way down is not a metaphor here. It is the procedural posture.
Why this is a conservation problem and not just a legal mess
This blog has argued before — in the context of algorithmic fairness, in the context of penalty structures, in the context of Arrow’s theorem — that formal impossibility results do not dissolve when you delegate the problem to a higher authority.2 They relocate. The higher authority faces the same structural problem the lower one did, expressed in the vocabulary appropriate to that level.
The Kalshi jurisdictional dispute is an instance of this. The underlying classification problem — is this a futures contract or a wager? — cannot be resolved by identifying the right classifier. Every candidate classifier (the CFTC, the federal courts, the state courts, the Ninth Circuit, eventually the Supreme Court) faces the same prior question: by what authority do you get to make this determination, and is that authority itself settled? The answer at each level is: not entirely, and see the level above.
The formal structure is identical to the problem of auditing a classification algorithm. You cannot audit an algorithm using the same algorithm — you need a meta-classifier that evaluates the classifier’s behavior. But the meta-classifier is itself a classification system, subject to the same questions about its design, its objectives, and its authority. Who audits the auditor? The question doesn’t have a stopping point. It has levels.
In algorithmic governance, this is a known problem and an active research area. In legal jurisdiction, it is called the constitutional structure of American federalism, which was designed with full knowledge that the question “who decides who decides?” does not have a final answer — and which assigned the Supreme Court the role of provisional last word, subject to revision by constitutional amendment. The design acknowledges the recursion. It doesn’t solve it.
What makes the Kalshi case interesting as a formal matter is that the recursion is compressed. All four levels are active simultaneously, in overlapping courts, with contradictory interim rulings already on the books. The system is not waiting for a higher level to resolve the lower one. It is running all levels in parallel, and the outputs are inconsistent. Tennessee says futures contract. Nevada says wager. Massachusetts says wager. The CFTC says futures contract and also says state criminal prosecution is inappropriate. The Arizona federal court says it is not sure it can say anything at all.
Friday: what happens when the classification doesn’t describe the activity but constitutes it — when calling something a futures contract is not a finding about what it is, but a decision about what it becomes.
With that, I leave you with this.
1 The “stunning concession” language is Arizona’s, from their briefing in the federal district court. The underlying procedural point is genuine: Kalshi’s preclusion argument, if accepted, would require the federal court to issue a ruling on the merits of the preemption question — which is what Kalshi wants — without technically enjoining the state proceedings — which is what the Anti-Injunction Act might otherwise prohibit. Whether this distinction survives scrutiny depends on how the court reads the Act’s exceptions. The Younger abstention doctrine, which counsels federal courts to abstain from interfering with ongoing state criminal proceedings as a matter of comity rather than strict statutory prohibition, adds a further layer. Younger abstention is discretionary; the Anti-Injunction Act is not. The court raised both.
2 The conservation of impossibility argument appears in this blog in several forms: in the context of Arrow’s theorem and aggregation (All Measurements Are Local), in the context of algorithmic fairness and penalty ceilings (The IRS Is Here to Help), and in its most general form in Can a Game Know Its Own Rules? The common thread: structural problems do not dissolve when delegated upward. They change address.
3 thoughts on “Who Classifies the Classifier?”
Comments are closed.