Your Simple Guide to Synth (SN50)
Predicting where a financial asset will move next is one of the hardest problems in both traditional and decentralized finance. Most forecasting tools produce a single number, a best guess of where a price will land. However, markets do not move in straight lines. They cluster in volatility, spike without warning, and produce fat-tailed outcomes that simple models fail to capture. Synth tackles this challenge by building a decentralized prediction engine on Bittensor that generates probabilistic price forecasts instead of single-point estimates. Specifically, it coordinates hundreds of competing AI models to simulate thousands of possible price paths for each asset, every hour.
What is Synth?
Synth is a decentralized financial forecasting subnet running as Subnet 50 (SN50) on Bittensor. It generates high-fidelity synthetic price path data for cryptocurrencies, commodities, and tokenized equities. Instead of asking one model to predict one price, Synth tasks hundreds of machine learning models with generating up to 1,000 simulated price paths per asset per request. Initially, miners produced 100 paths per asset. In November 2025, the network upgraded to 1,000 simulated paths, reflecting its push toward high-frequency trading capabilities. As a result, the output captures the full probability distribution of future price movements rather than a single estimate.
The approach relies on Monte Carlo simulations, a well-established method in financial modeling. Each miner builds and runs its own forecasting model that simulates realistic market dynamics, including volatility clustering, fat-tailed distributions, and sudden shifts in price.
Synth launched on the Bittensor mainnet in January 2025. Mode Network, an Ethereum Layer 2 built on the Optimism Superchain, developed the project and published a whitepaper outlining the technical design and incentive mechanism in detail.
The subnet initially focused on Bitcoin price forecasting. Since then, it has expanded to cover Ethereum (ETH), Solana (SOL), Gold (XAU), and five tokenized equities added in January 2026: the S&P 500 (SPYX), NVIDIA (NVDAX), Tesla (TSLAX), Apple (AAPLX), and Google (GOOGLX). This multi-asset coverage positions Synth as a broad financial forecasting layer within the Bittensor ecosystem.
How Synth works?
Synth coordinates three roles to produce and validate price forecasts across the network. First, miners run AI models that generate simulated price paths in response to validator requests. Second, validators evaluate those paths against actual market outcomes and assign scores. Third, the Bittensor protocol distributes rewards based on those scores.
Every hour, validators send requests to miners specifying an asset, a start time, a time increment of five minutes, a 24-hour time horizon, and a requirement for 1,000 simulated price paths. In addition to 24-hour predictions, Synth runs a second competition focused on one-hour high-frequency forecasting. Emissions are split equally between the two time horizons.
The scoring mechanism is where Synth’s core innovation sits. Validators assess miners using the Continuous Ranked Probability Score (CRPS), a statistical metric that measures both the accuracy and calibration of probabilistic forecasts. The system calculates CRPS on price changes measured in basis points across multiple time increments, including 5 minutes, 30 minutes, 3 hours, and 24 hours. As a result, miners must be accurate across both short-term and long-term horizons.
After each scoring round, validators cap the worst 10 percent of scores to the 90th percentile, subtract the best score from all results, and assign the 90th percentile score to miners who failed to submit valid predictions. The system then calculates a weighted rolling average over a 10-day window, with recent performance weighted more heavily through an exponential decay function. Finally, a softmax function converts these leaderboard scores into emission allocations. Consequently, consistently accurate miners earn proportionally more rewards over time.
Who is behind it?
Mode Network built and operates Synth. James Ross founded Mode Network and also leads Synth Data Co. The Synth whitepaper lists James Ross, Sam Hyatt, and Amber Shi as authors. Mode Network is an Ethereum Layer 2 network built on the OP Stack and part of the Optimism Superchain ecosystem. The company was founded in 2023 in the United Kingdom and received a grant in OP tokens from the Optimism Foundation to support its growth.
Mode Network’s broader strategy centers on what it calls AiFi, combining AI and decentralized finance. Synth fits directly into this vision by providing the predictive data layer that AI agents and trading systems need to operate effectively.
All Synth code is published on GitHub under an MIT license, allowing anyone to read, audit, or build on the codebase. The team maintains an active presence on X @SynthdataCo and communicates through the Bittensor Discord community.
Why it is valuable?
Synth addresses a fundamental limitation in financial forecasting. Traditional prediction models produce a single price target, ignoring the range of possible outcomes. This makes them poorly suited for risk management, options pricing, and portfolio optimization, all of which depend on understanding probability distributions rather than point estimates. Synth solves this by delivering ensemble predictions consisting of 1,000 simulated price paths per asset, providing full probability distributions, confidence intervals, and tail risk assessments.
The competitive structure of the subnet drives quality upward continuously. Hundreds of machine learning models compete on Synth, and only the most accurate forecasters earn meaningful rewards. The CRPS scoring mechanism evaluates miners against actual price outcomes, meaning their models must capture real market dynamics more effectively than established approaches like GARCH or Geometric Brownian Motion to rank highly. Because CRPS rewards calibration and sharpness equally, miners cannot succeed by simply predicting wide distributions.
Real-world applicability is already taking shape. The Synth API offers programmatic access to probabilistic forecasts for traders, developers, and institutions. Use cases include options pricing through Monte Carlo simulations, liquidation probability analysis for leveraged positions on perpetual DEXs, liquidity provision optimization for concentrated automated market makers, and integration with AI agents and trading bots. The API also supports Polymarket comparison data, enabling users to identify discrepancies between prediction market odds and probabilistic forecasts.
Synth also creates an open earning opportunity for data scientists worldwide. Anyone can register as a miner, deploy a forecasting model, and earn rewards based purely on prediction accuracy. Performance determines earnings, not affiliation or location.
The future of Synth
The Synth roadmap builds toward comprehensive portfolio-level forecasting. The team has completed support for Bitcoin predictions, expanded to multiple cryptocurrencies and equities, and is now working toward correlated asset simulations in Phase 4. This phase will enable miners to model how assets move together rather than independently. As a result, the output will become directly useful for portfolio construction and multi-asset risk management.
Integration with Mode Network’s AI infrastructure opens additional opportunities. Synth forecasts are designed to feed directly into Mode’s AI Terminal and on-chain trading agents for real-time risk assessment. The API and plug-in architecture allow direct integration with trading systems and on-chain AI agents, bridging decentralized forecasting data with existing financial infrastructure. Use cases in development include dynamic margin requirements for perpetual exchanges, real-time volatility surfaces for options platforms, and optimized liquidity ranges for concentrated DEXs. These integrations could make Synth a foundational data layer for AI-native finance across the broader crypto ecosystem.
Sources:
https://www.synthdata.co
https://docs.synthdata.co
https://github.com/mode-network/synth-subnet/
