Three orders. Friday afternoon. Nineteen dollars. Then seventy-four. Then eighty. Total exposure: one hundred seventy-three dollars. The position settled, roughly fifty-five times later, near nine thousand six hundred and eight dollars. The market had priced the outcome at one percent. The outcome arrived anyway. About twenty minutes after the last fill, a news outlet published the story that made the trade correct.
I have spent most of my working life reading order books for the shape of a lie. Usually the lie is small. A wash trade between two wallets sharing a funding source. A spoofed wall that evaporates when the book thins. A bot that front-runs a whale by three blocks and files it under "arbitrage." These are the ordinary bugs of a market whose rails anyone can inspect. You find them by grepping the chain, replaying the mempool, and watching where the value pools.
This was different. This was a wager on a real-world event, on a platform that does not run on a blockchain, on a contract that settles in dollars, placed by an actor the platform can name but the public cannot see. No transaction hash to grep. No mempool to replay. Only a private ledger, a compliance department, and a twenty-minute gap between the last fill and the headline.
The platform is Kalshi. The regulator is the Commodity Futures Trading Commission. The architecture is centralized, custodial, permissioned โ and that single design decision is the whole story. The amounts, the timing, the politics: all downstream of it. If you want to understand why a one-hundred-seventy-three-dollar bet is a systemic event rather than a curiosity, you have to stop reading it as a news item and start reading it as a system log.
The Market That Prices Belief
Prediction markets are not exchanges in the way most people understand the word. They are probability engines with a settlement layer bolted on. The mechanism is almost insultingly simple. A contract pays one dollar if an event occurs and zero if it does not. Therefore the price of the contract, expressed in cents, is the market's implied probability of that event. A contract trading at one cent is a market whispering "one percent." A contract trading at ninety-nine cents is a market shouting "near certain."
That equivalence โ price equals probability โ is the entire product. It is also the entire vulnerability. When you can buy a probability for a cent and redeem it for a dollar, the only thing standing between the market and a printing press is the accuracy of the probability. If the market is wrong by ninety-nine cents, the person who knows the truth has a riskless trade. There is no leverage to blow up, no liquidation to trigger, no funding rate to arbitrage. There is only the gap between what the market believes and what someone knows.
Kalshi operates one of these engines under federal supervision. It is a designated contract market regulated by the CFTC, which means it can offer event contracts to United States customers through fiat rails, with the blessing of a regulator that has spent years litigating whether such contracts are gambling, derivatives, or something genuinely new. Kalshi's answer, in effect, is: derivatives, supervised, legal. Its competitor Polymarket took the other road โ on-chain, permissionless, pseudonymous โ and paid for it with a CFTC settlement and a stretch of restricted access for American users.
Two architectures. Two trust models. One is fast, legal, and opaque. The other is slow, legally ambiguous, and radically transparent. The Kalshi incident is not really a story about a bad actor. It is a story about which of those two architectures can detect a bad actor before the money moves โ and which can only describe the theft after it is over.
I want to be precise about what this article is and is not. It is not a political column. It is an engineering post-mortem. When I reverse-engineered the 0x protocol's exchange contract in 2017, I learned that whitepapers are marketing and bytecode is truth. When I audited Curve's invariant equations in 2020 and found a precision-loss bug in the amplification coefficient that could be exploited during volatility, I learned that mathematical elegance does not guarantee safety. When I demonstrated in 2021 that a popular NFT project's minting function lacked access controls โ arbitrary token creation, treasury drained in seconds โ I learned that the crowd does not read the code, it reads the floor price. And when I dissected the reentrancy path in a lending protocol's liquidation contract in 2022, tracing the exact call-stack that turned a missing mutex into millions in losses, I learned that chaos rewards the calm. The lesson compounds across all of them: the trust model is the attack surface. So let us treat Kalshi the way I would treat any contract โ by asking what it assumes, what it records, and what it can prove.
The Pricing Engine That Cannot Lie About Probabilities
Start with the clean part, because it matters that the clean part is clean. The pricing mechanism is sound. Event contracts that settle at one dollar per winning share produce a direct mapping between price and probability, and that mapping is the most honest thing in the entire system. There is no yield curve to distort, no funding rate to manipulate, no liquidation cascade to engineer, no governance vote to capture. There is a single number: the probability the market assigns to a future. Everything else is plumbing.
This is why prediction markets are, in theory, the purest expression of information aggregation ever built. A liquid prediction market is a weighted average of everyone's beliefs, updated continuously, priced in real time, and settled in cash. It is the closest thing finance has to a truth machine. Economists love it for exactly this reason. It converts diffuse, private, contradictory opinions into a single public number that anyone can read and act on.
But a truth machine has a specific failure mode, and it is not the failure mode people expect. It does not break under volume. It does not break under volatility. It breaks under concentration of information. A truth machine is only as good as the diversity of its inputs. If a single participant knows the answer โ not believes, knows โ then the market stops being an aggregation of beliefs and becomes a transfer of wealth from the ignorant to the informed. The machine does not malfunction. It works exactly as designed. It prices the truth, because someone fed it the truth.
This is the first place where the crypto-native reader should feel a chill of recognition. The structure is identical to maximal extractable value. In MEV, a searcher with privileged information about pending transactions โ information that exists in the mempool before it exists on-chain โ extracts value from ordinary users by acting first. The information asymmetry is the product. The arbitrage is the mechanism. The victim is anyone who trades at the stale price. Sandwich attacks, backruns, JIT liquidity: all of them are the monetization of seeing the future a few milliseconds early.
In a prediction market, the same logic runs on real-world information instead of mempool data. The insider does not need to front-run a block. They only need to know a fact twenty minutes before the market does. The payout is the arbitrage. The loser is the counterparty who sold the contract at one cent, or the market maker who quoted it, or the casual trader who took the other side because the number looked like free money in the wrong direction.
Insider trading in a prediction market and MEV in a DeFi pool are the same bug wearing different clothes: both are the monetization of information asymmetry, and both are structurally legal until a regulator decides they are not. That sentence matters, and I will come back to it, because it is where the contrarian argument lives. The difference is not the logic. The difference is the enforcement layer wrapped around the logic. MEV is legal by default and policed only at the margins. Insider trading is illegal by statute and policed only when it is caught. Neither is prevented by the mechanism itself.
Let me put numbers on it, because numbers are where the argument stops being philosophical. If a contract trades at one cent and settles at one dollar, the gross return is one hundred times minus the cost, which in this case lands near fifty-five times on the actual stake after the market's spread and the platform's fee structure. A fifty-five-x return is not a trade. It is a lottery ticket that someone knew was not a lottery. In a well-functioning market, a one-percent probability should mean that one time in a hundred, the outcome occurs. When it occurs on the exact afternoon that three clustered orders were placed twenty minutes before the news, the probability of coincidence collapses toward zero. The market priced one percent. The information priced ninety-nine. The gap between them is the crime scene.
The Twenty-Minute Window
Now the forensics, because the pattern is where the evidence lives.
The record, as reported, describes three fills on a single Friday afternoon: nineteen dollars, then seventy-four, then eighty. The position settled near nine thousand six hundred and eight dollars โ a return of roughly fifty-five times, on a stake that would not buy a decent dinner in most of the cities where this trade was likely placed. Three details deserve a code auditor's attention, and a fourth deserves the suspicion that comes after the first three.
First, the escalating size. Nineteen, then seventy-four, then eighty. That is not the pattern of a confident insider dumping a large position into a thin book. It is the pattern of a tester. Small size first, to confirm the fill works and the contract behaves as expected. Larger size once the mechanism is verified and the liquidity is confirmed to absorb it. This is exactly how I probe an unfamiliar contract before I commit real capital to a live exploit: minimum viable transaction, observe the state change, read the receipt, then scale. The shape of the bet is the shape of someone checking their work.
Second, the timing. All three fills land within a narrow window on a Friday afternoon, and the public story breaks roughly twenty minutes after the last one. Twenty minutes is not a coincidence; it is a latency measurement. It tells you how far ahead of the news the actor was operating. It tells you that the information source and the execution were close enough together in time to plausibly be the same person or the same small group. In forensic terms, the window is the fingerprint. You cannot fake being early. You can fake almost everything else, but you cannot fake the timestamp of a fill relative to the timestamp of a headline.
Third, the total. Nine thousand six hundred and eight dollars is a strange number for a serious insider to accept. If you truly knew the outcome with near certainty, and you had access to real capital, you would bet six figures, not one hundred and seventy-three dollars. The modest size suggests one of two things: either the actor was capital-constrained โ a person with information but without a treasury โ or the actor was deliberately staying small to avoid tripping a surveillance threshold. Both readings are interesting. Neither is exculpatory. A small bet by a knowledgeable person is still insider trading. It is simply insider trading with better operational security.
Fourth, the pattern that the first three imply. The sequence โ escalating amounts, clustered timing, small absolute size โ is also consistent with copy-trading or coordination. If a sophisticated actor placed the first probe and one or two followers mirrored it, you would see a similar shape: a small initial fill, then increasing size as the signal is confirmed by others. The public cannot distinguish a single tester from a coordinated group, because the public cannot see the identity records. Only Kalshi can. And Kalshi has not said what it found.
This is the point where an auditor stops trusting the narrative and starts trusting the telemetry. The narrative says "suspicious trades." The telemetry says: three fills, escalating, clustered, small, twenty minutes ahead of the news. The telemetry is consistent with a single informed actor. It is also consistent with a small informed network. The difference matters enormously for enforcement โ one actor is a person, a network is a conspiracy โ and the public has no way to tell which it is looking at. That gap between what is knowable and what is knowable to the public is not an accident of this case. It is the defining property of the architecture.
A Ledger Only the Platform Can Read
Here is where the architecture decides everything, and where the comparison to on-chain markets stops being academic.
On Polymarket, every trade is a transaction. It has a hash. It sits on a public chain. Anyone with an indexer can reconstruct the full order history, the wallet graph, the funding sources, the timing, the size, the counterparties. The catch is pseudonymity: you can see the trade but not immediately the trader. So on-chain transparency gives you perfect forensic visibility and imperfect attribution. You can prove what happened. You struggle to prove who did it. The chain remembers the shape of the crime and forgets the name of the criminal.
On Kalshi, the situation is inverted. The platform does not publicly display trader names, but it retains the record of who placed every bet. The ledger remembers what the wallet forgets โ except here the ledger is private, and the only entity that can read it is the exchange itself. So Kalshi has perfect attribution and imperfect public visibility. It can name the trader. It cannot let the public verify that it did.
That asymmetry is the crux of the entire case, and it is worth slowing down to absorb. When a centralized venue investigates itself, the public is asked to trust the outcome of an inquiry it cannot audit. There is no block explorer for a compliance department. There is no proof-of-reserves for an internal investigation. There is no merkle root of the review process. The result is a trust model that reduces, ultimately, to "trust us" โ a phrase that sits awkwardly in an industry whose founding premise was the rejection of exactly that phrase.
And trust is precisely the resource that an insider-trading scandal consumes. The platform's entire value proposition is that its prices are fair โ that the one-percent probability means something, that the ninety-nine-cent probability means something else, that the number you see is an honest aggregation of belief rather than a trap set for you by someone who already knows the answer. The moment a market is revealed to have been priced by someone who knew, the fairness of every other price on the platform comes into question. A prediction market is a machine that converts public trust into prices. Insider trading is a withdrawal from that account, and unlike a blockchain, there is no way to audit the balance.
I have seen this exact fragility in a different form. When I audited the Curve contracts and found the precision-loss issue in the amplification coefficient, the danger was not that the bug would be exploited immediately. The danger was that the bug proved the invariant โ the mathematical promise at the heart of the product โ was not exactly true. The math was elegant and the implementation was close, and "close" is where the value leaks. A prediction market has the same relationship to its promise. The promise is "price equals probability." The reality is "price equals probability plus or minus the information advantage of the best-informed participant." When the information advantage is large, the promise is broken, and the product is a lie even though the code runs perfectly.
This is the first of the structural ironies I want to flag. Kalshi chose centralization precisely so it could be regulated, and regulation is supposed to protect market integrity. But the same centralization that satisfies the regulator also concentrates the investigative function inside the regulated entity, which means the integrity guarantee depends on the honesty of the very party whose product was just abused. The architecture that makes Kalshi legal is the architecture that makes it unverifiable. Code is law, but bugs are the human exception โ and here the bug is not in the matching engine. It is in the assumption that a self-policing venue can credibly police itself.
The Surveillance That Watched and Did Not Stop
Let me be fair to Kalshi, because fairness is a discipline, not a courtesy.
It almost certainly runs a surveillance system. A CFTC-regulated exchange does not operate blind; it has market-monitoring obligations, and it moved within hours to identify the suspicious trades and open an investigation. That speed is a signal. You do not identify three anomalous fills on a Friday and launch a probe by Monday unless you have order-level telemetry and the analysts to read it. The plumbing exists. The question is not whether Kalshi watched. It is why the watching did not prevent the trade.
There are three plausible answers, and they are not mutually exclusive. All three are engineering problems, not moral ones.
The first is threshold calibration. Surveillance systems are tuned. They have to be, because a system that flags everything flags nothing โ the alert fatigue problem that every security team knows. If the alert threshold is set to flag, say, fifty-thousand-dollar positions in low-liquidity contracts, then a one-hundred-seventy-three-dollar position never rings the bell. It is a rounding error in a sea of retail noise. The insider may have known this. The escalating-size pattern โ nineteen, seventy-four, eighty โ is precisely what you would do if you had a rough sense of where the bell was and stayed under it. You do not need to know the exact threshold. You need to know it is somewhere above you, and you need to stay below it.
The second is latency. Even a well-tuned system acts after the fact. By the time a flag fires and an analyst reviews it and a compliance officer decides it is worth escalating and a supervisor signs off on an inquiry, the news has broken and the money has settled. Surveillance is retrospective by nature. It generates evidence, not prevention. The twenty-minute window between the fill and the headline is, from the system's perspective, an eternity of latency. The system was never going to catch this trade before the news. It was only ever going to catch it after.
The third is the hardest to admit, and it is the one I find most instructive: the system was not built to catch this, because until recently nobody seriously believed insider trading in event contracts was a first-order problem. Event contracts were a novelty. The liquidity was thin. The political salience was low. Then the election cycle arrived, volumes exploded, and the same contracts that were once a curiosity became a real venue for real money โ and real information asymmetries. The surveillance did not fail so much as arrive late to a party that had already grown up. The threat model was written for a small market and inherited by a large one.
I have seen this exact failure mode in smart contracts, and it is the most common way protocols die. The audit is written for the code that exists at review time. Then the protocol grows, integrates new modules, adds a governance function, connects a bridge, and the assumptions that made the original audit valid quietly stop holding. The 2022 lending-protocol collapses I dissected were not caused by exotic bugs. They were caused by missing mutex checks that nobody flagged because, at audit time, the reentrancy path did not exist. The code passed its review. The world changed around it. The bug is rarely in the code you reviewed. It is in the assumption you stopped reviewing.
Kalshi's assumption was that its markets were too small to matter to insiders. The market outgrew the assumption, and the surveillance did not outgrow it alongside. This is not negligence in the ordinary sense. It is the ordinary failure mode of every system that succeeds faster than it hardens.
The Perez Anchor
If you want to predict how this ends, you do not need a crystal ball. You need a precedent, and there is one.
Earlier in this cycle, a case involving a Trump teleprompter operator โ Gabriel Perez โ resolved not in a criminal conviction but in a settlement: disgorgement of roughly one hundred seven thousand five hundred thirty-nine dollars, a sixty-five-thousand-dollar penalty, and a three-year trading ban. That is the quantitative anchor. It tells you the CFTC's enforcement calculus for this category of misconduct: return the profits, pay a fine that stings but does not destroy, accept a multi-year exclusion, and move on. It is a template, and templates are how regulators scale enforcement across a caseload they cannot try individually.
Two features of that precedent matter for the Kalshi case.
First, the disposition was civil and settled. That is the modal outcome for insider-trading-adjacent enforcement in a young market. It is cheaper for the regulator, faster for the respondent, and it avoids the burden of proving intent in a courtroom where the law itself is still forming. If I had to bet on the shape of the resolution โ and I try not to bet on anything without reading the contract โ I would bet on a settlement, not a prosecution. The uncertainty of the legal standard makes settlement the rational choice for both sides. The regulator gets a scalp without a risky trial. The respondent gets a fine instead of a felony.
Second, the penalty was calibrated to the profit, not to the harm. Disgorgement plus a fine that roughly tracks the gain. That calibration creates a perverse incentive, and it is the same incentive that haunts DeFi exploits: if the expected penalty is a fraction of the expected gain, and the probability of detection is low, then the rational strategy is to cheat and pay the fine. The fine becomes a fee. The fee becomes a cost of doing business. In 2021, I demonstrated how a clone project's minting function lacked access controls, allowing arbitrary token creation; the exploit was trivial, and the reason it existed was that the developers had priced the cost of the vulnerability at zero. Regulators who price insider trading at "give it back plus a fine" are making the same error. They are telling the market that the exploit is cheap.
This is where the quantitative anchor becomes a warning rather than a comfort. The Perez anchor tells the market what the price of getting caught is. When you publish a price, you invite people to decide whether it is worth paying. And a prediction market is a venue where people make exactly that kind of decision for a living. The enforcement community should assume that the anchor will be read as a price sheet, not as a deterrent. The only question is whether the price is high enough to matter, and the answer depends entirely on the size of the trade the next insider is contemplating. For a nine-thousand-dollar trade, a sixty-five-thousand-dollar fine is a disaster. For a nine-million-dollar trade, it is a rounding error. The anchor deters the small and not the large โ which is precisely backwards.
The Cost of Compliance and the MiCA Parallel
Now widen the lens. This is where the Kalshi case stops being a single platform's problem and becomes a sector's, and where a European reader should start paying attention.
The regulatory response is already escalating. The White House has issued a memo warning staff against profiting from non-public information, in the specific context of betting markets tied to geopolitical events. Lawmakers have opened inquiries into several other platforms. The CFTC is being pulled, case by case, toward a set of rules it has never clearly written. This is the classic pattern: a market grows faster than its rulebook, an abuse surfaces, and the regulator responds by tightening everything at once, because the alternative โ writing targeted rules for a category it does not fully understand โ is slower and politically harder.
I have watched this exact dynamic play out in Europe, and it is the reason I am skeptical of regulation-as-clarity. Europe's Markets in Crypto-Assets framework, MiCA, is the textbook case. It gave the continent apparent clarity โ a single licensing regime, defined categories, a path to compliance, a passport to operate across member states. What it actually did, on the ground, was impose reserve requirements and compliance costs that small stablecoin issuers and small crypto-asset service providers cannot absorb. The regime is not wrong in principle. It is wrong in its incidence. MiCA did not kill bad projects. It killed small ones. The compliance burden is a fixed cost, and fixed costs are regressive. A large issuer spreads the cost across billions in float and calls it a rounding error. A small one eats it alive and calls it a shutdown.
The same logic is now descending on prediction markets, and the mechanism is identical. If the CFTC responds to the Kalshi case by mandating surveillance systems, reporting obligations, and audit trails, the cost of those mandates will fall hardest on the smallest platforms โ the ones least able to afford them and most likely to be the honest, thinly-traded venues that never had an insider problem to begin with. Meanwhile, the well-capitalized venues will absorb the cost, raise their moat, and consolidate the sector. Regulation that was written to protect users will function as a barrier to entry, and the market will look cleaner because there are fewer players, not because it is fairer.
And here is the crypto-native irony that nobody in the enforcement chain seems to have priced: the regulatory tightening that follows this scandal will likely push activity toward the on-chain venues it was meant to discipline. Polymarket and its peers are pseudonymous, but they are also permissionless and, crucially, they do not carry the same compliance cost structure. If centralized prediction markets become expensive to operate, the marginal operator migrates. The rules intended to clean up the centralized venues may end up exporting their users to the decentralized ones โ a rerun of a movie we have all seen before, where prohibition pushes demand into channels the prohibition cannot reach.
I have a related concern about where this ends for the technology itself, and it connects to something I have argued for years about Layer 2. In my rollup work, I have watched zero-knowledge systems struggle under the weight of proving costs that make sense only when gas is expensive. The operators bleed when fees are low and the market is calm. The economics of compliance are the same shape: you pay a fixed cost continuously and recover it only in favorable conditions. Small projects do not survive the cold stretches, and the cold stretches are when the fixed costs bite hardest. So when I hear "regulation brings clarity," I translate it. Regulation brings fixed costs, and fixed costs bring consolidation. Clarity is for the survivors. The rest are the clarity.
The Information Circle
Return to the specific facts, because the structural analysis only earns its keep if it explains the particular case.
The winning position appears connected to a person who provides communications consulting to the media company that operates a major political social platform. The appointment at the center of the trade was confirmed on the same platform. The circle is small: a media company, a communications consultant, a confirmation post, a bet placed twenty minutes before the news. Every node in that chain is within one or two degrees of every other node. That is not a coincidence. It is the geometry of the vulnerability.
Small circles are the ideal environment for information leakage, and they are the hardest to police, because the leak does not require malice. It requires proximity. Someone in the room mentions a decision before it is public. Someone else, hearing it, does the arithmetic: the market prices this at one percent; I know it is closer to certain; the trade pays fifty-five to one. The distance between "overheard" and "traded" is a single app. No briefing document changes hands. No secret is sold. The information simply moves through the ambient air of a small network, and someone breathes it in and converts it to cash.
This is not unique to politics. It is the same dynamic as exchange listing leaks โ the token that pumps before the announcement, the wallet that accumulates the week before the integration, the anonymous account that posts a suspiciously accurate prediction an hour before the press release. I have audited enough of these to recognize the shape. The insider is rarely a mastermind. The insider is usually someone standing close enough to the information that acting on it feels like a technicality rather than a crime. They do not experience themselves as thieves. They experience themselves as people who happened to be in the right room at the right time.
That is precisely why insider trading is so hard to deter with penalties alone. You are not deterring a professional. You are deterring an amateur who does not think of themselves as a criminal โ who sees a one-percent market and a known fact and experiences the gap not as fraud but as free money. The psychology is closer to a bug than a heist. And bugs, as I keep insisting, are the human exception. The human exception is not a villain. It is a person who found a gap between what the rules say and what the room knows, and stepped through it because the gap was open and the money was on the other side.
The political dimension makes this worse, not better, and I want to be careful here because it is easy to slide from analysis into speculation. If the investigation concludes that the winning trades came from inside the information circle around a media company closely aligned with the administration, the matter stops being a market-integrity case and becomes a political one. At that point the resolution is negotiated in a different arena, with different incentives, and the market-integrity question โ the only question that matters to a prediction market's architecture โ is buried under the politics. I will not speculate about identities. That is not my job and it is not the useful part. The useful part is structural: a prediction market's fairness depends on the information circle being large and leaky relative to the size of the bets, and political information circles are small and tight while political bets are large and salient. The mismatch is the vulnerability, and it is architectural, not personal. You could replace every actor in this case and the vulnerability would remain, because it is a property of the shape of the network, not the character of the people in it.
Success Breeds Abuse
Here is the paradox at the heart of the sector, and it is worth stating plainly because it will not go away.
A prediction market becomes more valuable the more accurate it is. Accuracy attracts liquidity. Liquidity attracts attention. Attention attracts participants with information. Participants with information make the market more accurate. And the more accurate the market becomes โ the more reliably it prices the truth โ the more it diverges from the stale prices that ordinary participants trade at, and the more profit there is for anyone who can act on information before the market digests it.
In other words: a prediction market's success is the fuel for its own abuse. The better it works, the more it pays to cheat it. This is not a bug you patch. It is a property of any system that prices information, and it is the reason that insider-trading enforcement in mature markets is a permanent, never-finished war rather than a solved problem. Equity markets have spent a century building surveillance infrastructure, and insider trading still happens. The mechanism is not fixable because the mechanism is the product. You cannot have a market that prices information without having a market where information is worth money, and you cannot have information worth money without having people who want to capture it.
The crypto-native reader will recognize this too, because it is the same reason MEV has never been eliminated despite years of research and hundreds of millions in mitigation. The value being extracted is real. The extraction is structural. Every mitigation โ private mempools, batch auctions, encrypted order flow, intent-based architectures โ moves the extraction rather than removing it, because you cannot remove the information asymmetry without removing the information, and the information is the point. So the honest framing of the Kalshi case is not "a platform failed." It is "a platform discovered the cost of success." The market got good enough, and large enough, and politically salient enough, that the information asymmetry at its edges became worth exploiting. That is not a scandal. That is maturity, arriving with its invoice.
The practical implication is that the sector should stop pretending the problem is solvable and start treating it as manageable. You do not eliminate insider trading. You bound it, you price it, you make it expensive enough that the marginal insider decides the trade is not worth the risk. That is a threshold problem, not a moral one, and it is exactly the kind of problem an engineer can work on. The failure of the Kalshi case is not that someone cheated. It is that the threshold was set so low that cheating was cheap. Fix the threshold, and you have not fixed the bug, but you have raised the cost of the exploit. That is the most anyone has ever achieved in a century of trying.
The Contrarian Case: Maybe Centralization Is the Feature
Now let me argue against myself, because the evidence cuts both ways and a forensic analysis that only cuts one way is marketing.
The conventional crypto-native take is that Kalshi is the villain of this story because it is centralized. A decentralized prediction market would not have this problem, the argument goes, because everything is transparent. But look closer at what transparency actually buys you.
On-chain, the trades are visible. Good. But they are pseudonymous, so the trades being visible does not tell you who made them. And the information asymmetry that makes insider trading profitable does not disappear on-chain โ it simply becomes legal, because there is no insider-trading prohibition in a permissionless market. The person who knows the answer can trade on it openly, in public, and the chain will faithfully record the profit. Transparency does not prevent the extraction. It just documents it. And documentation, without attribution, is not accountability. It is a receipt with no name on it.
Meanwhile, Kalshi's centralization gives it something the chain cannot: identity records. It knows who placed those three bets. That means it can investigate, attribute, and โ if the regulator requires it โ disgorge. The on-chain venue can see the transaction but cannot name the trader, which is why the CFTC's on-chain enforcement actions have historically been clumsy and rare. The centralized venue is worse at transparency and better at accountability. The decentralized venue is better at transparency and worse at accountability. Neither is strictly superior. They are different failure modes, and the choice between them is a choice about which failure you can live with.
This reframes the whole debate. The question is not "centralized or decentralized." The question is "which failure mode do you prefer, and for which use case?" For a market where the abuse vector is insider trading, a centralized venue with identity records is genuinely easier to police โ the trade-off is that you must trust the policer. For a market where the abuse vector is censorship or counterparty risk, the decentralized venue is superior โ the trade-off is that you cannot police what you cannot attribute. Prediction markets sit squarely in the first category. Their dominant risk is the informed trader, not the censor. And against the informed trader, identity records beat transparency, because you can punish what you can name.
I want to push this further, because it connects to something I have been arguing about decentralized finance for years. Uniswap's V4 hooks turn the exchange into programmable Lego, and the same complexity that makes it powerful will scare off most developers. The complexity is the point and the problem at once. The same is true here. Kalshi's centralization is the point โ it is what makes it legal, fast, and accountable โ and it is the problem, because it is what makes it opaque and self-policing. You cannot separate the feature from the bug. Code is law, but bugs are the human exception, and the human exception is baked into every architecture, centralized or not.
The contrarian conclusion is uncomfortable for crypto natives, and I will state it plainly because it is what the evidence supports. The centralized platform may actually be the better venue for prediction markets specifically, precisely because insider trading is the dominant risk and accountability is the dominant requirement. The decentralized venue's transparency is a virtue for a market where counterparty risk dominates, but it is close to useless against the person who simply knows the answer. Transparency tells you what happened. It does not stop it from happening. And on a prediction market, the whole game is stopping the person who knows. If you want to build a prediction market that survives its own success, you need identity, and identity requires a center, and a center requires trust. There is no architecture that escapes the trade-off. There is only the choice of which cost to pay.
What to Watch
I do not write conclusions, because markets do not conclude. I write signals, because signals are what you can act on.
Watch the disposition. If the Kalshi investigation ends in a civil settlement modeled on the Perez anchor โ disgorgement plus a fine โ then the sector has learned that insider trading is a cost of doing business, and the incentive to cheat remains intact. If it ends in a criminal referral, the calculus changes, because the price of getting caught stops being a number and starts being a liberty.
Watch the rulebook. If the CFTC responds with surveillance mandates and reporting obligations, the compliance cost will consolidate the sector around the largest venues and push the margin toward the on-chain ones. That is not a prediction market story. That is every regulated market's story, told again, with the same ending: the small die, the large absorb, and the demand migrates to whatever channel is cheapest.
Watch the architecture. The real question is whether a prediction market can be built that prices truth without paying insiders for the privilege. I do not think it can, not entirely, because the profit is the point. But I think it can be built so that the extraction is visible, bounded, and priced โ the way MEV was eventually named, mapped, and partly internalized rather than eliminated. That is the realistic ceiling. Not a market without insiders, but a market where insiders pay a fair price for their information advantage.
And watch the ledger. The ledger remembers what the wallet forgets โ that is the promise of every market, centralized or not. The Kalshi case is a test of whether a private ledger can keep that promise without letting the public read it. I am not optimistic. Not because the people are bad, but because the architecture asks them to be good in the dark, and architecture is a poor substitute for verification. The twenty-minute window is closed. The three orders have settled. The only question that remains is the one the market cannot price: whether trust, once spent, can be earned back. Prediction markets will keep pricing everything else while they wait for that answer.