Subnet 50
Synth
Mode Network
Synth Subnet generates precise synthetic price data, capturing full probability distributions for AI-driven trading.
SN50 : Synth
| Subnet | Description | Category | Company |
|---|---|---|---|
| SN50 : Synth | Synthetic data for price forecasting | Predictive systems Data Pipeline | Mode |
The Synth Subnet harnesses Bittensor’s decentralised intelligence network to generate the most advanced synthetic data for price forecasting. Rather than relying on traditional price prediction methods that produce single-point estimates, Synth is designed to capture the entire distribution of potential price movements along with their associated probabilities, creating the most precise synthetic data available.
Within this network, miners are responsible for producing multiple simulated price trajectories that accurately represent real-world price behaviours, including characteristics such as volatility clustering and fat-tailed distributions. Their outputs are rigorously assessed using the Continuous Ranked Probability Score (CRPS), a metric that evaluates both the calibration and sharpness of predictions by comparing them against actual market movements.
Validators play a crucial role in scoring miners based on their accuracy across both short-term and long-term forecasts, with more recent performance carrying greater weight through an exponential decay function. Daily token emissions are distributed proportionally to miners’ relative performance, fostering a highly competitive environment that incentivises consistent precision in forecasting.
The Synth Subnet aspires to become the leading provider of synthetic price data for AI-driven trading agents, serving as an essential resource for options trading and portfolio management. By offering deep insights into price probability distributions, it aims to revolutionise data-driven financial decision-making.
The Synth Subnet is designed around a miner-validator system where miners submit their forecasts for future Bitcoin prices, and validators assess their accuracy after the fact. This ensures that rewards are distributed fairly based on performance.
Role as a Miner: Submitting Price Predictions
As a miner, your job is to generate simulated price paths for Bitcoin over a 24-hour period. Every forecast submission follows a structured set of parameters:
- Start time (t₀): Within 1 minute from the request
- Time horizon (T): 24 hours
- Time increment (Δt): 5 minutes
- Total number of increments (N): 288 (i.e., 24 hours ÷ 5 minutes)
- Number of simulated paths (Nₛᵢₘ): 100
This means that at each time step (tᵢ = t₀ + i × Δt), you will submit 100 simulated price predictions. One of the most commonly used models for generating these paths is Geometric Brownian Motion (GBM), but as the subnet evolves, more sophisticated techniques may be required.
Your success depends on accurately capturing real-world price dynamics, including volatility clustering and the skewed, fat-tailed nature of price movements. The better your simulations reflect reality, the higher you will rank.
How Validators Score Your Forecasts
To ensure fairness and promote independent, high-quality predictions, the Synth Subnet uses a systematic scoring methodology. This scoring system is structured in multiple stages:
1. Evaluating Your Forecasts
After the 24-hour forecast period, validators calculate your Continuous Ranked Probability Score (CRPS) at each time step (tᵢ). This score measures how well your forecasted price distribution matched reality.
The CRPS assessment focuses on two critical aspects:
- Calibration: Do your predicted price changes align closely with observed market movements?
- Sharpness: Is your forecast distribution tight and precise, rather than overly broad?
This evaluation is repeated for multiple time horizons—5 minutes, 30 minutes, 3 hours, and 24 hours—to assess your performance across short-term and long-term price forecasting. Your total CRPS score is the sum of these individual scores, providing a holistic measure of accuracy.
2. Aggregating Your Scores
Once individual CRPS scores are determined, they are summed to create your total unnormalized score. This ensures you are evaluated based on overall performance rather than a single time horizon.
3. Normalization for Fairness
To ensure fair comparisons, your total score is normalized using a softmax function. This has two major benefits:
- Fair Comparisons: The softmax function scales your score relative to other miners, ensuring rewards are distributed proportionally. A scaling factor (β = 0.001) prevents extreme outliers from unfairly dominating the reward pool.
- Encouragement of Independent Strategies: If multiple miners submit identical forecasts, they share the reward. This discourages copying and incentivizes originality.
4. Leaderboard Scoring & Performance Tracking
Your normalized score is incorporated into the leaderboard, which tracks both short-term and long-term performance.
- The leaderboard uses an exponential moving average (EMA) of past scores to balance recent and historical performance.
- At launch, the EMA half-life is 1 day, meaning recent performance has a greater impact.
- As the competition progresses, the half-life will increase, rewarding those who demonstrate consistent excellence over time while still allowing new miners to compete.
5. Reward Allocation Based on Performance
To amplify the rewards for the best-performing miners, the leaderboard scores are squared (raised to the power of 2). This magnifies score differences, ensuring that those with the most accurate and well-optimized models receive the largest share of emissions.
As the subnet matures, this exponentiation will increase, making distinctions between top performers even finer and pushing miners toward higher levels of accuracy.
The End Goal
Your success as a miner directly contributes to making Synth the most powerful source of synthetic price data for AI-driven finance. By continuously refining your models and improving prediction accuracy, you help build a market-leading platform that delivers high-value probabilistic forecasts for AI agents, options traders, and portfolio managers.
Awaiting Data