Precog and Bittensor: Redefining Crypto Market Intelligence
An overview of Precog and Bittensor's decentralized approach to crypto price forecasting
Get the best data-driven crypto insights and analysis every week:
Precog and Bittensor: Redefining Crypto Market Intelligence
By: Uriel Morone
Key Takeaways:
Bittensor is a decentralized network that incentivizes users to solve AI tasks by rewarding the best performers.
Precog uses Bittensor's infrastructure to create a marketplace for crypto price predictions, where forecasters compete for rewards.
Introduction
Earlier this month, Coin Metrics shared news about the launch of Precog, a cutting-edge subnet built on the Bittensor network. Built in collaboration with Yuma, a subsidiary of DCG, Precog aims to transform the landscape of cryptoasset price forecasting through decentralized machine learning and collaborative innovation. In this issue of Coin Metrics’ State of the Network, we'll dive into what Bittensor is, how its subnets work, and why Coin Metrics is excited to bring our data to the Bittensor ecosystem.
What is Bittensor?
Bittensor is a platform which uses blockchain technology to create decentralized networks tasked with solving a range of intelligent tasks. Traditional data sharing methods rely on centralized platforms, but with Bittensor new methods and types of analysis are possible which are distributed and resistant to manipulation and censorship, while retaining high throughput and accuracy.
$TAO tokenomics resemble that of Bitcoin. Only 21 million will ever be minted, but the emission schedule lasts 140 years, so the token price is quite inflationary (current value of 1 $TAO token is around $600). The token had a fair launch with no VCs, no investors, no premine, and Bittensor is not a company. The founders own less than 1% of the total token supply.
By leveraging the power of decentralized networks and machine learning algorithms, Bittensor empowers users to share and monetize their data in a trustless and autonomous manner. It uses a blockchain to coordinate rewards to users providing the best intelligence. This blockchain tracks the balances of users across many different subnets; which are networks designed to reward users for solving arbitrary tasks like AI natural language answers, graph optimization, protein folding, and market data prediction. These users can participate permissionlessly by paying a registration fee, incentivizing the best ideas from the broader community to offer their solutions to these tasks.
Coin Metrics is proud to integrate our Community data, which has been available as a core value since our founding, into a new subnet called Precog which will allow users with the best market knowledge to prove it while earning incentives for forecasting future price movements.
How do subnets work?
The purpose of a subnet is to coordinate several agents to solve an assigned task independently, identify the best solution according to public uniform rules, and ultimately surface data related to that task. As mentioned, the task can be extremely flexible. For example, in our Precog subnet the task revolves around forecasting the price of BTC. The end results are two forecasts: the exact BTC price one hour in the future and the max and min values that the price is expected to move during that hour. These forecasts are modeled at the discretion of contributors and the ones who perform best will receive the most incentive to produce future forecasts.
The following are a few examples to give a sense of the variety of challenges the subnet paradigm can address: There are subnet applications dedicated to responding to AI prompts (both image generation and natural language text), GPU Computing rental, sports betting, and more.
What they share in common is the tasks are accomplished by independent agents using their own solutions. In some cases the agents are competing to provide the best model, in others they are competing to provide the cheapest resources. However, in all cases the Bittensor structure enables a diverse set of solutions to a single problem and promotes the most successful.
Source: Bittensor Documentation - Understanding Subnets
A subnet is a network consisting of three types of agents. While you are likely used to the concept of Miners and Validators in different blockchain protocols, Bittensor uses these terms differently. In Bittensor, Miners and Validators do not secure the blockchain, they perform tasks within each subnet and their scores are written to the blockchain.
While these terms may have similarities to their usage in other contexts, it is best to initially understand Subnet Miners and Subnet Validators solely based on their definitions below. The three types of agent on a subnet:
Subnet Miner - A user which receives a request to perform a task and shares the results of that task back. In the Precog Subnet this will be forecasted value of BTC price and the interval it will trade between one hour in the future
Subnet Validator - A user which issues the requests to Miners and later ranks the performance of the Miners’ response from best-to-worst. In the Precog Subnet, Validators will compare prices forecasted one hour ago to the Coin Metrics reference rate during that same window and identify the miners which perform the best.
Subnet Owner - The user which registered the subnet, controls the hyperparameters, and controls the code which Validators run to identify the best Miners and reach consensus.
To align everyone’s incentives Bittensor uses the following system for emissions: Validators will rank Miner’s responses to their request according to an incentive mechanism pre-determined by the Subnet Owner. This mechanism can evolve over time as the owner updates the subnet. The Miners will receive “weight” from the Validators according to how they have been ranked, which in turn will determine their share of the TAO token emissions.
Validators also act autonomously, but they are judged according to the Yuma Consensus algorithm. There are a few components which incorporate results over time, but broadly speaking Validators will be scored highly based on how closely the weights they assign to Miners are in consensus with the weights assigned by the other Validators. In this way Validators can act autonomously and the ranking of Miners can be decentralized but all are incentivized to promote Miners that best fulfill the task of the subnet, because that is how the other Validators are expected to rank.
How does Precog Work?
The Precog Subnet will focus on market data predictions, powered by Coin Metrics’s powerful Community API. Miners will be tasked with two outputs:
The price of Bitcoin 1-hour from the prediction time (Point Forecast)
A price range equal to the min and max value of the price during the 1-hour following prediction time (Interval Forecast)
These outputs are generated by the Miners every 5 minutes and evaluated by Validators an hour later. The focus is to use the decentralized structure of Bittensor to produce refined, easily digestible, and applicable data points about future price forecasts at higher time resolution than what is possible in traditional exchanges.
Traditional derivatives markets provide some insight into market belief of future price movements, but the raw data consists of a ladder of contracts at different strike prices, expiration dates, and rules. Quantitative Analysts and mathematical models are required to transform the data on a derivatives market into meaningful probabilities of future price.
The Precog Subnet in contrast is not a market at all, it focuses specifically on producing valuable intelligence. The task of processing market data to answer “whats the most likely price” is offloaded to the Miners, who can solve the problem using the methods they find most effective. For data consumers, the output of the subnet are two actionable data points that can be utilized with minimal post-processing: What’s the price going to be? and how much will it vary?
Stay up to date with Precog
We have plenty of ideas on how to utilize the data output by the subnet and extend what we can do once the community is established. We are currently in the process of developing a dashboard so the public can view how forecast performance evolves over time. Stay up to date with the latest developments at:
Network Data Insights
Summary Metrics
Source: Coin Metrics Network Data Pro
Over the past week, Ethereum and Avalanche C-chain saw daily active addresses rise by 10% and 15% respectively, while Bitcoin's declined by 3%. On Solana, USDC supply surged 51% to 9B, as Trump-related memecoins drove heightened activity and an influx of liquidity, pushing Solana's total stablecoin market cap above $10B for the first time.
Coin Metrics Updates
Follow Coin Metrics’ State of the Market newsletter which contextualizes the week’s crypto market movements with concise commentary, rich visuals, and timely data.
As always, if you have any feedback or requests please let us know here.
Subscribe and Past Issues
Coin Metrics’ State of the Network, is an unbiased, weekly view of the crypto market informed by our own network (on-chain) and market data.
If you'd like to get State of the Network in your inbox, please subscribe here. You can see previous issues of State of the Network here.