I keep coming back to a question that sounds boring until you sit with it. Where, physically and legally, is the money I put up to bet on a market outcome. Not the market I'm betting on, the collateral itself. Because when a prediction market platform goes wrong, it almost never goes wrong at the level of the bet. It goes wrong at the level of the escrow. And the two dominant models for holding that escrow fail in completely different ways, which means the right amount of money to leave on each one is different too.
On one side you have the regulated clearinghouse model, the Kalshi shape. Your dollars sit in segregated accounts at a bank, held by a designated contract market that answers to a US regulator. On the other side you have the on-chain model, the Polymarket shape. Your stablecoins sit in a smart contract on Polygon, and settlement happens when an oracle reports the outcome and the contract pays out. Same activity from the outside. Wildly different failure surface underneath.
What can actually go wrong with segregated funds
The clearinghouse model borrows its safety story from decades of futures regulation, and most of that story holds up. Segregation is real. Your collateral is supposed to be walled off from the operating company, so if the platform itself burns cash and goes under, the customer funds are not supposed to be part of the bankruptcy estate. That is the whole point of the structure and it is genuinely protective.
But segregation is a legal promise, not a physical law, and I've watched enough of the FTX aftermath to know that the gap between how funds are supposed to be held and how they are actually held is where people lose everything. The failure modes I actually worry about here are not exotic. One is commingling, where the wall between customer money and house money turns out to be more of a suggestion. Another is the bank holding the segregated account, because segregated funds still sit at a bank, and banks fail, and deposit insurance caps are low relative to what a serious trader parks. A third is operational, the platform freezes withdrawals during a dispute or a regulatory event, and your money is safe in the legal sense while being completely inaccessible in the sense that matters when you want to move it.
So the clearinghouse model protects you well against the venue simply losing money, and protects you poorly against the venue, or its bank, becoming a place you cannot get your money out of on the timeline you need.
What can actually go wrong with a smart contract escrow
The on-chain model deletes the counterparty in one sense and replaces it with three new ones. There is no company holding your funds that can commingle them, which genuinely removes the FTX-style failure. The contract holds the collateral and the contract pays out, and you can read the contract. That is a real improvement and I don't want to undersell it.
The replacement risks are just less familiar. The first is the contract itself. Code has bugs, and an escrow contract that has never been drained is not the same as an escrow contract that cannot be drained. Historically the largest single-event losses in this space have come from contract exploits, not from anyone running off with the money in the traditional sense. The second is the oracle, the mechanism that decides what actually happened so the contract knows who to pay. If resolution runs through a dispute-and-vote system, then a market that is genuinely ambiguous, or one that someone is motivated to contest, can settle in a way you think is wrong, and there is no clearinghouse to appeal to. You were right about the world and the oracle disagreed, and the contract paid the oracle's answer.
The third one is the quiet one. Your collateral is a stablecoin, which is itself a claim on a separate issuer's reserves. A depeg is a full-portfolio event on an on-chain venue. It does not matter that your bets are all resolving in your favor if the dollar-shaped thing they are denominated in trades at ninety cents while you are trying to exit. You took a currency risk you never chose to take, stacked on top of the market risk you did choose.
A framework for capping exposure per venue
Here is how I actually size this, and it starts by refusing to treat the two models as one asset class with one risk number. The trick is to ask what has to be true for the money to survive, and put less on the venues where more independent things have to hold.
- Count the independent points of failure. For a clearinghouse venue that is roughly the platform's solvency, the bank's solvency, and the platform's willingness to let you withdraw. For an on-chain venue it is the contract, the oracle, and the stablecoin issuer. More independent failure points means a lower cap.
- Cap on-chain exposure to what you'd accept losing to a single bad resolution or exploit. Treat any single contract as a thing that can go to zero for reasons that have nothing to do with your read of the market. That framing keeps the number honest.
- Hold stablecoin collateral in the issuer you would still trust during a panic, and keep a mental haircut on it. If you would not want to be the last person redeeming that stablecoin, do not park size in it.
- For clearinghouse venues, size to the withdrawal freeze, not the insolvency. Ask how much you can afford to have locked and untouchable for weeks during a dispute, because that is the more likely bad day, and set the cap there.
- Split across models, not just across venues. Two on-chain platforms sharing the same stablecoin and a similar oracle design are far more correlated than they look. Real diversification means one of each model, because they fail for unrelated reasons.
The rule of thumb I keep is that the venue holding your money the most abstractly gets the smallest cap. A regulated segregated account is concrete and legible and can hold more. A smart contract escrow denominated in a stablecoin is three abstractions deep, so it holds less, and I keep enough off it that a contract exploit or a depeg is annoying rather than ruinous.
None of this tells you which side of a market to take. It tells you how much you can afford to be right and still get hurt, which is the part people skip. When I'm tracking where collateral sits across venues inside Blockcircle, the custody model is the first column I look at, before the odds, because the odds only pay you if the escrow is still standing when the market resolves. Know how your money is held before you decide how much of it to hold there.