In Late 2024, Beton Sports LLC engaged the product of subnet-41, aka: Sportstensor with regard to its ability to generate alpha in one of the most difficult markets in the world, the high volume low margin sports betting market.

The results were profound; We produced the highest net return in our syndicates history by multiples.

This report will act as an unbiased and forensic accounting for the efficacy of Bittensor as a product that claims to facilitate the creation of useful machine intelligence. It seems when incentive mechanisms are exponential in nature, they attract miners who can vastly outperform the standard quo.

What is Beton LLC?

Beton LLC is a private group of LPs that operate a betting syndicate, which is just a fancy way of saying we pool capital to bet into sports betting markets. We use data science and market experience to allocate capital and generate a return for our LP’s and the LLC takes a fee for this on profits.

In general over the years we have generated about a 3% return on investment, on average, with out best sport, baseball returning a phenomenal 5.4% per dollar bet.

These rates of return are considered elite in the business. Simply this means that for ever dollar we bet into markets we have an Expected Value of 3% (EV) and Return of 3 cents on the dollar bet (ROI).

Beton LLC often engages with originators who claim they are +EV in sports and gives them a cut of the profits if they win. About 3 in 1000 win. This is the true rate of success in our field. When we engaged Sportstensor’s ensemble learning model we agreed to create a test account to test if their return on investment was appreciable enough to become a customer and then go ahead and pay them for their services in NBA, EPL and MLB.

The Forensic Accounting System

To account for the product we created an account on Pinnacle Sports LLC, we do not mind sharing the account #. It is only a testing account amongst hundreds that we operate. We injected $900 USD to start and decided to keep our bets to a flat and static ratio of 3% of the bankroll per bet so that the efficacy of the model is tested in a fair and sterile way with no “spiking” of bets in order to pad the stats.

Every day we looked at the probabilities that Sportstensor provides via their API and made judgements from their ensemble model consisting of 3 NBA bettors that are clearly winning. If the implied probability of the bet, say 1.8 to 1 or a 55.56% win probability was higher than the odds given by the bookmaker, say 2.0 or a 50% chance of winning, we would bet this because we are generating 20 cents per dollar bet of EV. The bets are usually conducted at about 10 AM PST.

The bets are made and observed in the field. When the games are done we report to our LP’s the result of the bet, the number of bets made, the account balance and account number as well as the betting receipt as can be seen below.

In this way we forensically track the daily performance. If we make money we scale the size of the capital up proportionally so that the flat betting amount equates to the agreed upon flat bet percentage. If subnet-41 was any good, we ought to make money.

The results are observed by 127 unique LP’s.

The Results.

On January 4th we commenced betting. The following image is our first tracking report and starting amount.

On Feburary 13th we stopped due to the NBA all star break and decided to report our findings on a sample size of over 300 bettable games (a significant number)

It turns out that on pure betting operations, we generated a return of approximately 3.6x compounding up from $50 CAD bets to $160 CAD bets. We noticed that the intelligence had an intriguing spread of favorites and underdog bets and often acted contrarian to the market and won while still respecting strong teams and not making the common mistake of overfitting and taking too many underdogs.

At an average bet of $80 USD over 150 bets we generated $2540 USD in profit representing a per dollar bet return of approximately 18.41% return on the dollar bet.

This constitutes the best return on investment we have ever seen in 9 years of betting and engaging at least 50 originators over this time in NBA betting. We conclude that whoever is mining sportstensor has, in effect, completely dominated the NBA betting market in ways that the common originator has not factored into the market.

We can only conclude that Sportstensor has invented new market intelligence that is novel and unique. Bittensor in effect has shown that incentive undeniably creates intelligence.

Caveats

There are a few points we should make that downplay the result.

• 350 total bettable games is not a lot of games in our opinion we would like to see this performance keep up through 1000 games to be more certain
• We did hedge about 10 out of the 175 matches out so maybe 15% ROI is more realistic but that is just our long experience in betting NBA that we use on top of the model output.
• We filtered out matches that were not worth betting. (odds of 1.2 or less or 5 or higher) as its just silly to risk money on a random outcome for almost no money or take risks on teams that tank on purpose.
• Maybe the results won’t be this good for MLB. We will continue monitoring.
• We are concerned that the team will give the data away to too many entities, thereby eliminating the edge. We will monitor this closely for edge erosion. We would prefer to pay them much more and forbid others from accessing the data.
• When you reach a certain size, the bookmakers will no longer let you bet, so there is a limit to the market access this intelligence can generate without sophisticated account management.
• We would like to see the product attract more than 3 good miners. Its not enough. We would like to see an ensemble of 100 miners who win creating a truly dominating multi armed bandit market.

Conclusion

Working with the subnet-41 team has been very pleasant. The team are avid developers who deeply respect advice and feedback on their odds construction and have been open to the experience of pro bettors who know the markets inside and out.

When looking for subnets to invest in, its very important to ask them for their products and do tests in the field.

• Does the product show signs of improvement over the cutting edge? Yes it does.
• Does it have an addressable market that scales? Yes it does.
• Is the team fast to respond and humble? Yes they are.
• Is the incentive mechanism exponential thereby forbidding questionable intelligence from forming on their protocol? Yes it is.

We believe now that subnet-41 is on the path to machine dominance and will remain paying customers for their services and start scaling up our operation to bet on their data. We will buy the token with a portion of our returns on a daily basis.

Follow us on X: https://x.com/beton_ai

Follow sportstensor at: https://x.com/sportstensor