It’s been less than a month since when we first registered Precog, and we’d like to thank our miners, validators, and the overall Bittensor community for your eagerness to work and grow with us.
Early Highlights:
Miner Growth / Participation
- We’re currently sitting at 245 miners and have seen tremendous and steady growth in interest:
- Precog is unique amongst subnets in that miners can be successful with just a laptop and some straightforward analytical work (this will and has become more difficult with increased competition driving better outputs).
- PreCog leverages the Coin Metrics community API, which is free and robust, leveraging 148 different data metrics with some receiving updates in 1 second frequencies.
Ecosystem Feedback and Support
- After speaking with over a dozen validators and OTF we’ve received background knowledge, critiques, praises, and had a few doors slammed in our faces proverbially speaking.
- We’re aware there is a historical aversion to price forecast subnets and as a critical aspect of our roadmap are looking to show how we can learn and overcome these historical challenges. More below on this.
Things We’ve Learned and What We’ve Shipped
- We had a period of about 5 days (see the registration chart above from 1/22 – 1/28) where only 33 of 120 active miners were receiving emissions due to a validator with large weight having an older code version running. We’ve since learned how to troubleshoot this faster and better educate validators day 1.
- We wrote a research paper on Bittensor and Precog to Coin Metric’s subscriber list of 20K+ crypto researchers and users. Since the research report we’ve seen about 25 new miners come to Precog per day.
- Merged an update for miners that cache’s our base miner to make data fetching less intense.
- Added wandb integration for validators to allow the forecasts / outputs of the subnet to be stored and eventually displayed to the ecosystem.
Short Term Roadmap
In the near term we’d like to focus on two main areas we feel will be most impactful for future growth and set us up for success under the impending dTAO changes:
Iterating on our initial design to prevent exploitive participants and reward authentic miners:
We’re going to keep our exact methodology intentionally vague here, as the point of limiting exploitive mining is to create processes that can’t be gamed; however, we’d also like to share some high level ideas we’re testing currently:
- Something basic like a longer-moving average. This reduces the chance of getting a forecast right at random.
- Requiring Miners to share their exact calculations with us, possibly through a Github the Validators would pull and run themselves.
- Running an internal statistical model on miners to determine the authenticity of their forecasts over time (i.e. randomness in forecasts is less rewarded than forecasts that show a consistent methodology in data).
- Using statistical similarity to incentivize forecasts that are unique (and successful), while dampening weight for forecasts that tend to be very similar to others.
Data exports and visualizations to benchmark Precog’s output value:
- To bring the outputs of Precog to life, we are currently implementing a dashboard and leaderboard of sorts to show where value is being derived and to also help understand if and where incentives are flowing.
- In a dTAO world we need to greatly accelerate how the ecosystem observes our outputs and why they are valuable in an easy to digest manner.
Future Ideas / Longer Scope Work
We have many ideas for innovative directions to develop this subnet, and grow alongside the Bittensor ecosystem. Some are more immediately accessible and some represent more ambitious long term possibilities:
Expanding the current task to a larger universe of tokens or different genres of assets: TAO, NFTs, memcoins, majors, DeFi, etc.
- The less liquid the markets and the more signal is obscured by noise, the more value Precog produces by consolidating information and forecast models.
Feed the outputs generated by Precog through an API that can be programmatically accessed by Coin Metrics clients.
- Other forms of data access may be possible internally on the Bittensor EVM, using smart contracts as well.
A wider array of market data types to be forecast.
- In time the subnet tasks may expand beyond simply predicting price, to identifying the most valuable indicators and models that predict miner behavior. Allowing users in the future to take what the subnet has learned and use it for their own varied purposes.
Monetization of Precog
- Coin Metrics has been licensing crypto data since 2017 to some of the largest institutions in the space. The outputs from Precog have synergies with our existing business model, and can create a flywheel for success under dTAO.
- We’re also exploring ways of incentivizing Miners to share the most valuable data types for market forecasts. The idea here is using miners to help discover which data points are most highly correlated, or contain the most information, about future market movement. This may ultimately be a more monetizable output than the predictions themselves.
Stay Updated
Keep in the loop of all things Precog! You can find us on:
