A wire item crossed my feed on election night: prediction market participants are pricing a higher probability that Democrats take both chambers of Congress. That was the entire claim. No venue named. No ticket size. No timestamp. No settlement clause. No bid-ask.
I have spent months of my life auditing smart contracts and building liquidity models for regulators. That sentence is not a data point. It is a rumor with a chart attached, and the chart is missing its axes. When a market claim arrives without a print, a depth, and a resolution source, the correct treatment is not skepticism β it is classification: this is not information, it is an index entry pointing at information that was never collected.
Here is what a probability of 19 cents actually contains, and why almost none of it is the thing the headline says it is.
Prediction markets come in two mechanical flavors, and the difference is not cosmetic.
Polymarket runs a central limit order book on Polygon, collateralized in USDC, with matching executed off-chain and settlement pushed on-chain through an optimistic oracle. Kalshi is a designated contract market under the CFTC β USD margin, named clearing, position limits, and a regulator that can and has told it what it may not list. In September 2024, Kalshi won the right to list election contracts after a federal court rejected the agency's attempt to block them. One venue settles in a token on a public ledger. The other settles in a bank account under a rulebook. Same question, different instrument.
That distinction matters more than usual here, because Polymarket's own regulatory history shapes who is in the book. In January 2022 the company settled with the CFTC for $1.4 million over offering off-exchange binary options to US persons, and blocked US IPs. The crowd whose wisdom gets cited in election coverage is, by construction, not a sample of the American electorate. It is a sample of people who hold USDC on Polygon and are legally permitted to use it.
So what is a probability in this market? It is a marginal price on a finite book. Nothing more.
If twenty thousand dollars of aggressive buying walks a contract from 12 cents to 19 cents, the honest reading is "thin book," not "sentiment shifted seven points." A market order for size eats through the asks; the last trade you see quoted in a news item may represent the top of the book, which is precisely where prices are least informative. Quoted mid and executable price are different numbers, and the gap between them is the only part of the contract that reflects reality rather than narrative.
Volume makes this worse, not better. During the 2024 presidential cycle, Polymarket's headline market cleared several billion dollars in notional turnover, and reporting in late 2024 raised credible questions about how much of that was genuine directional flow versus incentivized round-tripping. Volume is cumulative and historical. Liquidity is instantaneous and finite. The two can move in opposite directions for weeks, and the metric that gets published is the one that gets incentivized.
Then there is the clause. "Democrats take both chambers" sounds like a single question. It is at least four: majority of seats in the House and Senate; majority including the vice president's tiebreak; net seat gain above a threshold; or a specific seat-count band. A Senate that lands at 50-50 resolves differently depending on whether the contract's resolution language mentions the tiebreak at all. Two contracts with different resolution text are two different assets, and quoting them against each other is a basis trade, not a consensus.
I learned this the hard way in 2018, during three months auditing 0x Protocol v2 for my graduate work. I filed seven edge-case pull requests, and not one of them concerned the happy path. Every serious defect lived in the dispute window β the moment where the system's interpretation of its own rules gets tested by an adversarial party. Optimistic oracle designs work on the same premise: someone posts a bond, someone else either accepts it or disputes it. The risk you are pricing in an election contract is not only who wins. It is whether the resolution source is unambiguous at the instant of settlement, and whether anyone is economically motivated to contest it.
Now the piece almost nobody models: carry. USDC locked as collateral in an event contract does not earn yield. That is an opportunity cost that must be embedded in the price. On a sixty-day contract at a 5% risk-free rate, the hurdle is roughly 0.8 points of edge just to break even against sitting in T-bills. A contract trading at 19 cents is not simply a 19% belief. It is a 19% belief minus a carry charge minus a slippage charge minus a resolution-risk discount.
And the cost of capital you would use to lever that position is not discovered. Borrow curves on the major lending markets are set by governance parameters β a slope voted in by token holders, a base rate adjusted at intervals β and they respond to utilization ratios rather than to any observable price of time. The rate you pay to express a view in DeFi is an administrative decision wearing the costume of a market price. In a bear tape that matters, because collateral is expensive and patience is cheap.
Finally, the timestamp. Odds are a time series. A snapshot without a print is not evidence β it is a claim about a moment that cannot be located. In 2022 I modeled the Terra collapse as a liquidity cascade rather than an ideological failure, and the number that made the report citable was that roughly sixty billion dollars of stablecoin value left the system inside forty-eight hours. Direction alone was obvious for about six hours before magnitude became obvious. Direction without magnitude is entertainment. Magnitude without a timestamp is folklore.
The prevailing view is that prediction markets are superior information aggregators β cleaner than polls, faster than pundits, self-correcting because money talks. I think that thesis is structurally incomplete, and the blind spot is the composition of the marginal trader.
Three filters determine who can set the price. A capital filter: you need stablecoin on a specific chain. A jurisdictional filter: US persons have been excluded from the deepest venue. A market-maker filter: a meaningful share of resting quotes come from professionals managing inventory and hedging exposure elsewhere, whose quotes reflect delta management rather than conviction. Stack those three, and the marginal price is often a hedge price, not a forecast. The market is not telling you what it believes about November. It is telling you what it costs to lay off risk right now.
Which redirects attention to the number that actually carries signal: the spread between the regulated venue and the offshore one on an identically worded question, and the depth at which that spread survives an order of a given size. If the gap is three points and it costs 2.5 points of slippage to close it, the market is not pricing a sweep. It is pricing the cost of arbitrage. Liquidity doesn't vote β it prices.
Watch three variables over the next cycle, and ignore the headline probability until you have them: depth at a stated price, the exact resolution clause including its dispute window, and the venue's regulatory status at the moment of settlement. Everything else is a quote without a counterparty. If the settlement source can be contested, the number was never a forecast. It was an offer β and somebody will send you the bill for taking it.
