Subnet 50: Synth
Synth is Bittensor Subnet 50. The Synth website describes the project as synthetic asset price-path data generated through a decentralized network of AI models, while the Synth README source frames the subnet around probabilistic price forecasting.
What Synth Provides
The README describes a Bittensor subnet for synthetic price data and probabilistic price forecasting. Instead of asking miners for a single future price, the subnet asks miners to submit ensembles of simulated price paths that represent possible future movements and uncertainty.
At the subnet level, Synth turns forecast quality into a miner-scoring signal. Validators request price-path forecasts, later compare those forecasts with realized market movements, and publish weights that feed into Yuma Consensus.
Forecast Distribution Context
Synth’s core forecasting context is distributional forecasting. The README says miners submit multiple simulated price paths for supported crypto, equity, and commodity assets, with the paths intended to represent the miner’s view of possible future price movement rather than one point estimate.
That makes the miner output closer to a synthetic market-data distribution than to a direct trading call. A useful submission needs to express uncertainty, including the shape and spread of possible price paths, because validators later compare the submitted distribution with realized market movement.
The project website uses similar framing by describing synthetic asset price-path data from a decentralized network of AI models. Synth is therefore best read as a forecasting and synthetic-data subnet: miners generate probability-aware market paths, and validators transform forecast quality into Bittensor weights.
Forecast Horizon Context
The Synth README source
describes miner tasks across two forecast timeframes: a 24h forecast and a 1h high-frequency
forecast. Both ask miners for ensembles of simulated price paths, but they test different parts of
the same probabilistic forecasting problem.
The 24h task asks for a longer path over a full-day horizon. The README describes validator
scoring across multiple price-change intervals inside that day, including shorter checks and the
full 24-hour change. That makes the daily task broader than a single closing-price guess: the
forecast path has to remain useful across several slices of the day.
The 1h high-frequency task uses a shorter horizon and finer price-change intervals. It tests
whether a miner’s simulated paths can describe fast market movement over minutes rather than over a
full day. This distinction is why the article should treat Synth as distributional forecasting
across timeframes, not simply as one generic market prediction task.
The README also describes per-timeframe rolling windows for miner scores. The shorter horizon
generates observations more quickly, so its recent-performance window is shorter than the longer
24h window. For readers, the important point is that Synth compares forecasts in the context of
the horizon being tested instead of flattening every request into the same evaluation rhythm.
The supported asset set can change as Synth adds markets, but the horizon boundary remains the stable reader-facing distinction. A shorter task and a longer task can include overlapping market types while still measuring different forecast behavior.
This horizon context helps explain the subnet’s value as synthetic price-path data. The useful artifact is not just a final price estimate; it is a set of probability-aware paths that can express uncertainty at both short and longer horizons.
References: Synth README source, Synth website
Scoring Context
The README identifies the Continuous Ranked Probability Score as Synth’s main evaluation rule. CRPS is useful here because it evaluates probabilistic forecasts, not only whether a miner guessed the correct direction. The source describes lower CRPS values as better because they indicate a forecast distribution that better matches the observed price movement.
Synth’s validator scoring also uses recent forecast history rather than a single isolated request. The README describes rolling windows for the different forecast timeframes, so recent performance continues to matter while older observations age out of the scoring window.
This scoring shape makes Synth different from subnets that reward storage, inference throughput, or liquidity commitments. Its weight signal comes from repeated probabilistic forecasts across market assets and timeframes, with validators rewarding miners whose simulated paths remain well calibrated against later price movement.
Miner and Validator Roles
Miners generate probabilistic forecasts for supported crypto, equity, and commodity assets. The README describes miner submissions as multiple simulated price paths across short and longer forecast timeframes.
Validators evaluate the submitted ensembles using the Continuous Ranked Probability Score, a scoring rule for probabilistic forecasts. The README describes lower CRPS values as better forecasting performance, with validator scores averaged over rolling windows before weights are set on chain.
On-Chain Identity
Live SN50 subnet data is available on TaoStats. The source-backed forecasting and scoring details in this article come from the Synth website and README rather than from live identity fields.
Relationship to Yuma Consensus
Subnet 50 uses Yuma Consensus to convert the forecast-quality weight vectors that validators submit into the emission shares distributed to miners and validators within the subnet each tempo. The linked documentation describes how validator weight submissions are aggregated into consensus weights for each miner registered on the subnet.
In Synth’s context, validators request probabilistic price-path ensembles from miners, later compare those forecasts against realized market movements using the Continuous Ranked Probability Score, and translate rolling-window CRPS averages into weight vectors for the subnet. The Emission documentation describes how those consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Synth (SN50), that sequence changes how readers should interpret probabilistic price-path forecast examples and CRPS scoring outcomes.
In localnet, Synth-compatible miners and validators can be developed and tested in an isolated environment. Localnet forecast evaluations and emission outcomes do not represent production subnet performance.
On testnet, Synth-compatible price-path forecast submissions can be exercised in a shared, non-production network. Testnet forecast scores and validator weights are separate from mainnet subnet state.
On mainnet, Synth (SN50) is the live production subnet where miners submit probabilistic price-path ensembles and validators score those forecasts against realized market movements to determine real Bittensor emissions. The Synth repository describes the mechanism that applies on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A forecast score or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Reader Boundary
Subnet 50 Synth should not be read as generic Bittensor subnet documentation, a trading signal service, or proof that one point price guess wins rewards. It names one subnet’s probabilistic price-path forecasting market on netuid 50 (Understanding Subnets, Glossary: Netuid).
CRPS Scores Ensemble Paths Against Realized Movement
The Synth README describes validators scoring miner ensembles with the Continuous Ranked Probability Score, with lower CRPS values indicating better forecast calibration (Synth README source).
Rewards therefore follow distributional forecast quality rather than directional guesses alone.
Rolling Windows Separate 1h and 24h Horizons
The same README describes 24h and 1h forecast tasks with per-timeframe rolling score windows
(Synth README source).
Each horizon keeps its own recent-performance boundary instead of one flattened leaderboard.
Validator Weights Still Flow Through Yuma Consensus
Subnet 50 uses Yuma Consensus to convert validator weight submissions into emission shares each tempo (Yuma Consensus, Emission).