Every time I explain why one prediction market feels bottomless and another moves half a cent when you place a modest bet, I end up drawing the same little diagram. The difference is almost never the traders. It is the pricing engine sitting underneath, quietly deciding how much your order moves the price and how much edge is left on the table once it does. If you trade these markets, or you are trying to read signal off of them, the mechanics of that engine are worth more than any single price print.
Most of the early on-chain and academic prediction markets ran on an automated market maker, and the specific one worth understanding is the logarithmic market scoring rule, or LMSR. It is the design Robin Hanson wrote up in the early 2000s, and it quietly became the default for a decade because it solves a real problem. In a thin market for some obscure question, there is often nobody on the other side of your trade. A traditional order book just sits empty. LMSR gives you a counterparty that is always willing to quote, at a price, no matter how few people are paying attention.
What LMSR is actually doing
The mechanism is simpler than the name suggests. The market keeps a running count of how many shares have been bought for each outcome. Call those quantities q_yes and q_no. There is a cost function built on those quantities, and the price you pay for the next share is just the change in that cost function when you buy it. Buy yes shares and q_yes goes up, which pushes the cost function up, which raises the price of the next yes share and lowers the price of the next no share. The prices for all outcomes always sum to one, so they read cleanly as probabilities.
The important consequence is that the price is a smooth, deterministic function of net demand. You are not matched against another trader. You are walking up a curve. Small orders barely move you. Large orders move you a lot, and the cost of a big position is the integral of that rising price, not the flat quoted price you saw before you started. This is the same slippage idea that shows up in constant-product AMMs on decentralized exchanges, just derived from a different cost function. If you have ever watched a large swap eat its own price on a DEX, you already have the intuition. LMSR is that behavior applied to probabilities.
The liquidity parameter, and why it is the whole ballgame
There is one knob that controls everything about how the market feels, usually written as b. It is the liquidity parameter, and it sets how much money it takes to move the price a given amount. A large b makes the market deep. Prices barely budge, big bettors can get size on without wrecking the quote, and the curve feels almost flat near the middle. A small b makes the market thin and jumpy, where even a small bet swings the implied probability several points.
Here is the tradeoff that b encodes, and it is the part people miss. The sponsor of the market, whoever seeds it, is on the hook for the market maker's losses. The maximum the sponsor can lose is bounded, and that bound scales with b times the log of the number of outcomes. Bigger b means a deeper, more pleasant market, and it also means the sponsor has posted more capital at risk. Smaller b caps the sponsor's downside but hands traders a market that lurches around on tiny flow. So the depth you experience as a trader is really a statement about how much the sponsor was willing to subsidize.
That bounded loss is the elegant part of the whole design. The sponsor knows the worst case before the market opens. They are effectively paying a fixed, known subsidy to guarantee there is always a quote and to nudge traders into revealing what they believe. In exchange they eat some expected loss to informed traders. For research markets, internal forecasting inside a company, or long-tail questions nobody would otherwise make a market in, that trade is often worth it.
Reading the edge off the curve
Once you see the market as a curve rather than a book, a few practical things fall out. Here is roughly how I think about any AMM-based prediction market before I trade it or trust a print off it.
- Check the depth before you believe the price. A market quoting 62 percent on a thin b can be dragged to 58 or 66 by a single motivated bettor. The number is real, but it is soft. Size your read to the liquidity behind it.
- Price your full order, not the top quote. Because you walk up the curve, the average price you pay on a large position is worse than the price you saw. Estimate the fill across the whole size, not the marginal share.
- Watch for the subsidy running thin. As traders push the price toward one extreme, the marginal edge for the sponsor's maker shrinks and the market gets harder to move further. Late in a lopsided market, the curve is steep and your slippage is ugly.
- Separate flow from information. On a thin AMM, a big trade and a small trade can move the price the same visible amount if b is set aggressively. The price change tells you about demand, not always about new information. Do not confuse the two.
Why the big venues moved to order books
For all its elegance, LMSR has a structural weakness that matters at scale. The sponsor is always subsidizing the market, which means someone is always paying to keep it liquid. That is fine for a thin research question. It is a bad deal once a market gets popular enough that real traders would happily make markets against each other for free. At that point you are burning subsidy to provide liquidity that the crowd would provide on its own.
A central limit order book flips the economics. Instead of a sponsor eating bounded losses, you have market makers posting bids and asks and competing on spread. Liquidity gets deep because it is profitable to provide, not because someone funded it. The venue collects fees instead of paying subsidies. For any market with genuine two-sided interest, that is strictly better, which is why the largest, most liquid prediction venues run on order books rather than AMMs. The price discovery is sharper, the spreads are tighter, and nobody is footing a subsidy bill.
The order book does demand something the AMM did not. It needs enough participants on both sides at the same time to function. In a quiet market you get a wide spread or an empty book, exactly the failure the AMM was built to avoid. So the honest picture is not that order books won and AMMs lost. It is that they solve different regimes. Order books dominate wherever attention and volume are high enough to sustain competitive making. AMMs still earn their keep in the long tail, the niche and newly created markets where there is no crowd yet and you need a guaranteed quote to bootstrap one.
The practical takeaway is that the pricing engine is a lens on the edge available to you. On an order book, your edge lives in the spread and in reading order flow against other makers. On an AMM, it lives in the shape of the curve and in the sponsor's willingness to subsidize depth. When we pull prediction-market signals into Blockcircle alongside the whale and disclosure feeds, the first thing we check is which engine produced the print, because a 60 percent on a deep book and a 60 percent on a thin AMM are not the same claim about the world. Know the engine, and the number finally tells you what it is worth.