Introducing: Prediction Market Atlas What most people see in prediction markets is only the surface. The categories now shipping include futarchy, distribution markets, multiverse markets, opinion markets, attention markets, decision markets, and info-finance. Each has its own pricing logic, regulatory lane, and audience. PM Atlas maps that landscape across 108 platforms, 12 mechanisms, and 90 concepts, from industry structure down to the actual products being built. A way to track who is building what, how these markets differ, and where the next primitives are emerging. pmatlas.xyz

the biggest unpriced risk in a prediction market is its own volatility. on screen it's a clean probability. on the maker's book it jumps on every headline, then a fixed date drags it to 0 or 1, ready or not. and a new model fit on kalshi data found where that vol lives. it peaks at 50/50 and again right at the deadline. so the maker was never short the price. they're short a belief vol surface that spikes exactly where they're most exposed.

Introducing Blanket! Blanket helps small businesses find relevant @Kalshi markets for hedging risk and running promotions. Describe your business risks and it will help identify the Kalshi markets that fit. Blanket is a research product and will be open sourced after testing.

You’ve gotta be kidding me

Kalshi pays APY on locked collateral, Polymarket mostly does not. if longshot fat tails are behavior, that gap means nothing. if they are financing, the dead money venue prints the bigger yes premium and it grows with time to resolution, not with how wrong the crowd is. nobody has run the split and it won't be clean, the venues differ on more than collateral. my call anyway: by january 1st the yes premium comes out bigger where the money sits dead.

the underpriced layer here isn't the rails, it's the balance sheet. pull the house out and its risk lands on whoever makes the market. one injury reprices every leg and that maker eats the tail. anyone can buy the matching engine, surviving the sharps is the part that isn't for sale.

There was never a market for what an asset is worth in a future that has not arrived and Multiverse markets are that market! I wrote a piece on how they work and where they break. The odds were never the point, the world was.

Been a bit quiet these past few days. I lost one of the closest people in my life. It was the first time I’ve lost someone who meant that much to me and it was a strange experience. The thing I’ve been stuck on ever since is how you’re supposed to balance life in a way that leaves you with as little regret as possible when someone you love is gone. The more I think about it, the more I feel there’s always going to be some regret. Maybe that’s just part of it. Strange experience. Back on my feet. Time to get back to building.

a mechanism x chain map of the prediction market landscape, and it tells you exactly where the space actually is versus where it thinks it is 1/ it's still overwhelmingly binary. binary markets dominate every chain, 15 on solana, plus base, arbitrum, bnb, and a long tail everywhere. this is the prototype phase everyone building the same yes/no primitive on a different chain 2/ now look at the rows that barely have entries. continuous and distribution markets, 2. scalar, a handful. multiverse, a couple. the hard stuff is empty precisely because it's hard the other tell is that infrastructure and tooling is the second most populated row, 13 on solana, 10 on polygon. builders clearly sense the real value is in the rails, not another front-end. i believe that instinct is right solana is the center of gravity by a wide margin, with base quietly picking up the consumer and information market builders the category is wide but shallow. tons of binary clones, a thriving infra layer, and a nearly empty frontier where the continuous, scalar, and multiverse mechanisms live if you can, start building in the rows everyone else left blank

The part of the HIP4 spec I keep rereading is the bond economics. 500k hype slashable for poorly defined markets or wrong settlement, a six month lock and no unstaking until every market you deployed has settled. And that basically makes the deployer an underwriter of resolution quality, the resolution layer priced as capital instead of promised as process.

Polymarket shows you one parlay price and charges you another. the one it shows comes from the help page, a combo pays by multiplying the leg probabilities together. but the one it charges comes from a dealer auction you never see and the dealer does not use that math. with a 40¢ leg and a 50¢ leg, the help page says 20¢ but the fills say more. I measured the difference across 300 real fills, every leg rebuilt at its price the moment the combo filled. 285 of 300 landed above the product of their legs, even at ask prices. the median fill pays about 22% less than the multiplication implies, and simple two leg combos floor around 1.11x the product. one exact score leg never traded above 3.7¢ in its whole window and its combo filled at 9.4¢, a price that cannot happen in coherent math. there is also a fee in no documentation, roughly 5% of notional scaled by one minus the price, sitting right in the fill data. to be fair, part of the gap is earned. correlation is real work to price and a dealer carrying a parlay to settlement is underwriting a real tail. the problem is the interface teaches independence math the dealer never uses, so users read the price as the odds when it is the odds plus the house's whole opinion of them. the dealer prices correlation, the interface teaches multiplication. the gap between those two stories is the real fee, and nobody sees it until asking. all of it is public data, the method and dataset reproduce from the venue's own endpoints. the next generation of venues has to price correlation in the open and whoever shows the joint math honestly turns this seam from a fee into a market.

The worst World Cup final ever…

Been calling passive PM makers underwriters since june and someone finally measured it. Palumbo reconstructed the passive maker book on @kalshi NFL moneylines and the P&L comes from outcome risk carried into settlement more than from the spread, about $29M held through the whistle. LPing a binary venue is selling insurance with the premium set by accident. makers get paid for liquidity. underwriters get paid for carrying the result. binary venues sell the first story and run the second.

very interesting paper that i am just seeing now, guess it flew under the radar here on prediction market X because it comes from draftkings and has uncomfortable regulatory implications i do agree with the take though, a lot of the algorithmic strategies are more akin to underwriting than market making the paper itself is also pretty short so definitely worth a read if you missed it

Either you lock in for 2-5 years or You clock in for the rest of your life

Polymarket quietly turned its parlay button into a dealer auction. you tap a combo and a price appears, but underneath the makers get 400ms to quote it and the one who wins can still hand it back inside a second. that second part is last look, the most fought over privilege in fx, and it now sits under a sports parlay. so the book prices the legs and a dealer prices the combination. the industry that swore everything would be an open book rebuilt the phone call.