An Interview With Leo Chan from Sportstensor

“Sportstensor is collective intelligence leveraging a Darwinian decentralized network and incentivization to foster competitive excellence in sports predictions. We believe this novel approach generates predictive insights superior to those of traditional methods such as sportsbooks odds and institutional trading desks.”

“I was first introduced to Bittensor at the end of 2023 when there were still only 32 subnets. Having seen AI explode in popularity and interest during that time, and being in crypto since 2017, I found myself naturally gravitating towards what the intersection of both industries looked like which is how I found the TAO token. I delved into the deep, dark and often frustrating rabbit hole of mining TAO, which is how I met my co-founders. I connected with both of them separately while mining different subnets but brought everyone together as a way to expand our mining knowledge and the rest is history.”

“The new incentive mechanism isn’t too dissimilar to the previous in how we score predictive quality, the differences are mostly in execution and architecture. Over the last year, we learned a lot about building the best incentive mechanism possible for our use case and the biggest (probably the hardest) lesson was about value leakage. While we were incentivizing good outputs from our miners, the noise was also getting a piece of the pie. No amount of patches or refactoring of our existing code sealed the value leakage. Our new incentive mechanism solves for that completely, controlling the value (alpha) distribution to be the first subnet to crack a completely anti-dilutive incentive mechanism. It took a hard realisation and a foundational shakeup in how we built the subnet for us to get to this point.”

“Being able to create markets is important for us to incentivize information that people actually want. If a sports trading fund wants accurate probabilities of an obscure Finnish ice hockey league, we can make the market with Polymarket and direct incentives towards it for traders to build models and discover edges. Now imagine a hedge fund willing to pay millions for accurate weather forecasts to trade agricultural commodities or investment firms needing accurate geopolitical information to hedge portfolio risk during elections or policy changes. This gives us the ability to service institutional clients leveraging our partnership with Polymarket.”

“Almanac is an information incentivization layer for prediction markets. To the user/miner we are a prediction market front-end using Polymarket liquidity and infrastructure. We did this specifically in partnership with Polymarket but also to save time in building out a whole prediction market from scratch, so we can launch in a reasonable amount of time. Our target users are sharp traders, the ones who have an edge and are already trading through prediction markets. We’re giving them a reason to trade through Almanac, feed us their edge and get rewarded for it in potentially multiples more than they would just trading on prediction markets.

Building a prediction market is no easy feat, even with Polymarket serving as a large majority of the infrastructure. We wanted to take our time to make sure it is functioning as it should with the utmost care in security and stability when facilitating millions of dollars worth of transactions. With that said, we couldn’t keep everyone waiting for too long so December 1st is the slated beta launch of Almanac.”

“With a 1% trading fee on Almanac that all goes back to alpha buy backs, we are capitalizing on and optimizing for volume first and foremost as a revenue generator. In the short to medium term roadmap, we’re building meta-models from the miner outputs as we’ve done in the past, to then deploy into the markets and trade — once again, returning all profits to alpha in buy backs. Lastly, B2B sales with institutions or entities who would pay for incentivized information.

This creates a flywheel where miners provide useful information and earn alpha, we put the information to work generating revenue that goes into alpha, miners earn more which increases competition. Repeat.

This is where the value leakage solution really shines. Our alpha is only distributed to miners producing real signals that can generate revenue which all goes back to the alpha token. For example, let’s say on Almanac we have a mix of both miners and normal users, and we make $1000 in fees for the day. Let’s assume miners contributed $600 towards those fees. Then we’ll pay out $600 in alpha incentives to miners, and use the associated $600 in fees to buy back the alpha token. The other $400 in fees is used to pay out normal traders while $400 in miner alpha emission gets burnt. Any surplus miner incentives that aren’t required at that time are recycled for later use. This system then creates a closed-loop, anti-dilutive flywheel.”

“Esports is a booming segment of the sports betting industry as a whole, estimated at roughly $2.8 billion in volume per year. This is expected to have a double digit CAGR into 2030. It is a relatively new sporting category compared to baseball or basketball but with each match generating a couple million data points, it is the perfect environment for machine learning and data science to be applied. Probabilities and odds are soft, not many traditional sportsbooks know or even care to price them properly so there is an incredible amount of opportunity to have the most accurate predictions in this space.

When we decided to tackle this industry, we knew we needed to partner with the very best data source to exist today. Grid is a rare company where the founders recognized the potential of esports data long before the rest of the industry caught on. Grid is the official data partner to some of the largest esports titles in the world including League of Legends, Valorant and Counter Strike — having been granted exclusive data rights from the very servers of the games themselves. To put that into perspective, it’s like having direct access to NFL’s player tracking chips or the New York Stock Exchange’s order flow. No intermediaries, data straight from the source.

The partnership came about because Grid’s CSO and co-founder had been following Sportstensor’s progress through our Head of AI and when we got to chatting, the synergy revealed itself organically — data and information are the backbone of both machine learning models and prediction markets. Signing them on as our esports data partner and miner helps us continue driving value back to our alpha token.”

“The fundamental ethos of Bittensor miners is that incentives on subnets are there to be earned, it is up to the miner to determine if it is worth their time or not. Sportstensor has been incentivizing sports intelligence since June 2024, giving Bittensor miners the opportunity to try at an industry they have no experience in to earn rewards for their machine intelligence. Almanac is simply the front-end of the subnet that will open that same opportunity up to the world. It is only through presenting Almanac to professional sports bettors, hedge funds and international trading desks have we been able to garner interest and commitment to using the platform. We tried a more technical onboarding approach using Bittensor’s btcli solution but were met with nothing but silence. Not many people talk about this side of building a subnet but signing up real companies as miners or clients is extremely difficult. The complexity of Bittensor doesn’t help our case. We’ve gained a lot of experience speaking to these people which led to us designing our new incentive mechanism and Almanac.

This is a problem we’re solving that many if not all Bittensor subnets now face. Break out of the Bittensor bubble and reach real world adoption or fade into obscurity. We must abstract the complexities away and have people use Bittensor without realizing they are using Bittensor. As veterans of the ecosystem, we frame the question differently: It’s not what Bittensor can do for the subnet, but what the subnet can do for Bittensor.”

“Any sharp traders who’ve used both will probably tell you that prediction markets are far superior. The whole business model of most sportsbooks these days is built on losing bettors and they will do anything to optimize for that. They encourage losing bettors to keep betting, entice new bettors with bonuses and free money, limit accounts that appear to have an edge, trade against their users, or ban accounts that are consistently winning.

Prediction markets (at least the ones built on crypto rails) are open markets that welcome any traders from the casual bettors to the professional funds that deploy models built by their 20-man quant team. The information that flows through these markets is the business model. Evident by NYSE parent company ICE investing $2B into Polymarket, the information and data from prediction markets are extremely valuable to many different industries. When you heavily restrict the information from flowing (such as by sportsbooks) then you lose the value and truth.

Prediction markets already incentivize information through market dynamics but they are highly inefficient in optimizing for signals. We see a real opportunity to make the markets even more efficient in that respect starting with sports to prove the incentive mechanism. Once proven, this framework extends naturally to other prediction categories such as crypto, politics, weather etc. This scalability is why prediction markets represent the future beyond traditional sportsbooks, as infrastructure for aggregating collective intelligence across any domain.”