You Get Banned for Winning

From a Beyond Finance community call, in conversation with Harry Crane, Philip Maymin and Iosif Gershteyn of Djinn, Subnet 103.

---

Note: Since this article was first published SN103 Djinn has been de-registered

---

The sportsbook runs on a quiet rule most people never hit. Lose, and you're a customer they'll keep forever. Win, and you're a problem they solve. Prove you can pick winners consistently and they limit what you're allowed to stake, then close the account, and the edge you spent years building is worth nothing the day it's proven.

Harry Crane, a statistics professor and one of the three people behind Djinn, put it flatly when the team came on one of our Beyond Finance calls to explain what they're building. "Once you're profitable in sports betting, you become dispensable." The books have no use for a winning bettor. They only want losing ones.

So picture two people who will never meet. One genuinely knows how to beat the market and has been shut out of using it. The other is one of the hundreds of millions of dormant or unlimited betting accounts sitting idle around the world, held by someone with no edge at all, still perfectly free to place a bet. Knowledge with no way to act, and the ability to act with no knowledge. Djinn, Subnet 103 on Bittensor, is built to introduce them.

Split the knowing from the doing

Djinn calls itself the Information and Execution Subnet, which is drier than the idea deserves. The idea is an unbundling. Normally the person who knows which bet to place and the person who places it are the same person, bundled together by default. Djinn pulls them apart.

Iosif Gershteyn, who carried most of the explanation, reached for Uber, and it's the cleanest way in. Uber didn't invent the car or the passenger. It separated owning a vehicle from earning by driving one, and matched the two in real time. Djinn does that to information. On one side are the people the team calls the geniuses, the sharp bettors who know how to win. On the other are the ones the books treat as idiots, tolerated precisely because they lose, so they're left free to bet. "Djinn is the uber of this information execution network," is how Iosif put it. It doesn't much matter where the bet itself lands, FanDuel, DraftKings, a local book, because the network only checks that the odds were real and live at the moment the signal was sold.

The genius sells the signal, the idiot places the bet, and the design is that both come out ahead.

But why would anyone give away an edge?

The obvious objection came up straight away. If you can pick winners, why hand that to a stranger instead of just betting it yourself?

Because you already are, right up until you can't. Harry's answer is that every sharp bets their own edge for as long as the books let them, and that window is always closing. He described the pattern from experience. You open a fresh account, place five bets, lose all five, and you're banned anyway, because the book has profiled the way you bet and decided you're on the smart side of the line whether these five landed or not. Winning is a tell. And even the biggest operations, the syndicates that turned sports betting into fortunes and football clubs, hit the ceiling of what any single account will take. The edge is real. The room to use it keeps shrinking.

So the knowledge needs a way out of the person holding it. But that opens a harder problem than it solves.

The problem underneath the problem

Say you want to buy a winning signal from someone you've never met. How does that work without one of you getting robbed? You'd have to trust them to hand over a genuine, live bet rather than something worthless. They'd have to trust you to pay. And worse, anyone sitting in the middle who can see the signal, the network itself included, could quietly place it first, or sell it on to the next buyer before you've moved. Information is the one asset that's copied the instant it's seen.

This is where it gets interesting, and where Philip Maymin, who handles the cryptography, walked us through what they've built. Follow it slowly, because the abstract version sounds like hand-waving and the specific version doesn't.

Prove you win, reveal nothing

Start with the miners. Their only job is to confirm that the odds a genius claims were genuinely available at the time, at those prices or better. They never touch money and never hold a secret, so they don't have to be trusted at all. And to stop a miner noticing which game is being asked about and front-running it, every real question is buried in a crowd of fakes. Each query, as Philip put it, "is populated with at least 100 and sometimes 1,000 decoy lines," and the miner has to answer for all of them, live or not live, one after another, with no way to tell which one anybody actually cares about.

The validators handle the signal itself, and here the trick is that no single one of them ever sees it. They work together using multi-party computation, a way for several parties to reach one answer without any of them seeing the inputs, to confirm the real pick sits inside the set of acceptable ones without any single validator ever learning what it is. The genius's signal is encrypted so it can only be opened on the buyer's own device, and only after they've paid. Nobody in the middle sees the bet. Not the miners, not the validators, not Djinn.

Then the part that made the room go quiet. Because every signal is checked this way, a genius builds a track record that can be proven without ever being revealed. You can show, cryptographically, that someone has been picking winners for months, while never exposing a single one of the picks. Philip's word for it was "magic," and for once it wasn't overselling.

This is also the answer to a question people rightly ask about anything on Bittensor: why does it need to be decentralised at all? Harry gave the reason without being asked. If one company sat in the middle verifying these signals, that company could see them first and use them first. The only way to make "nobody can steal your bet" actually true is to build it so that no single party is ever in a position to. The decentralisation isn't decoration. It's the thing that makes the product possible.

Where the real-world money comes in

Most subnet tokenomics run on emissions, the network paying its own participants. Djinn's argument is that the money flowing through it comes from outside that system entirely, from an industry Harry sizes in the trillions. The network takes a small fee on each signal, half a percent of what's being staked, and the team has said it will use that fee to automatically buy the token back and burn it. Emissions pay the miners for their work; the value that rewards holders is meant to come from real betting flow, not from printing more of the token.

So that's the design. What's already real, and what's still a plan? The buy-and-burn was announced on our call and, at the time, hadn't gone live or even been posted publicly yet. The first fully automated, end-to-end signal, a real bet moving through the whole system without anyone stepping in, had happened only that week. This is a subnet at the very start of proving itself, not one showing you a finished machine.

And the team is unusually straight about its own weak spot. When the conversation turned to why so few people understand what Djinn does, Iosif named it himself. The hardest thing they've faced "surprisingly wasn't the technical challenges, it was the marketing and messaging." They can build a market that keeps a secret from everyone, themselves included. Explaining it to the rest of us is the part they find hard. That isn't a Djinn problem. It's most of Bittensor.

What it's really about

Strip the sport away and the shape underneath is bigger than betting. Djinn is a way for someone who knows something valuable, but can't act on it, to sell that knowledge to someone who can act, without either trusting the other and without the knowledge leaking on the way. Sports betting is just the first place they've pointed it, partly because the team came out of that world, and partly because the ban makes the pain so sharp. They're careful not to run ahead of themselves, the way PayPal started life tied to eBay before it became anything else. But the engine works anywhere the same split exists, anywhere the person who knows and the person who can act are two different people.

Most of what we build online still assumes those two are the same person. What changes when knowing and doing can be bought and sold apart?

https://x.com/djinn_gg https://www.djinn.gg https://x.com/imgpf https://x.com/harrydcrane https://x.com/pmaymin