Subnet 18: Zeus
Zeus is Bittensor Subnet 18 (SN18), operated by Orpheus-AI, with its codebase published in the Zeus repository. The subnet coordinates AI-driven global environmental forecasting: miners compete to produce the most accurate predictions of weather and climate variables across the Earth’s surface, evaluated against authoritative reanalysis data after ground truth becomes available. The goal is to push weather forecasting accuracy beyond the current state of the art by applying competitive incentives to the problem.
How the Mechanism Works
According to the Zeus repository, validators issue forecasting challenges that ask miners to predict environmental variables across a global 0.25-degree grid. The subnet runs two parallel forecasting horizons: a short-term challenge covering the next 48 hours, and a long-term challenge covering up to 15 days ahead. Miners submit their predictions and validators reveal scores after ground truth becomes available from ERA5 — a continuously-updated global reanalysis dataset maintained by the Copernicus Earth observation programme.
Scoring uses area-weighted error metrics (Root Mean Squared Error and Mean Absolute Error) that account for spherical grid geometry, giving appropriate weight to grid cells at different latitudes. Rankings aggregate over a rolling window of recent challenges rather than any single submission, ensuring that miners demonstrate consistent performance over time.
The reward structure follows a winner-take-most model: the top-ranked miner captures the large majority of weight for each challenge, with remaining weight distributed logarithmically to runners-up. Weight vectors from validator scoring feed into Yuma Consensus, distributing emissions via Dynamic TAO.
Participating as a Miner
The Zeus repository describes miners as producing global weather and climate forecasts in response to validator challenges. Their economic role is to develop or deploy AI models capable of predicting atmospheric variables accurately across both short and long time horizons. Because scoring aggregates recent performance rather than rewarding individual submissions, miners must maintain consistently high accuracy to earn significant emissions. The winner-take-most structure means that marginal improvements in forecast accuracy translate directly into a much larger share of rewards.
Participating as a Validator
Validators on Zeus issue challenges, collect miner forecast submissions, and reveal accuracy scores once ERA5 ground truth data is published, following the workflow documented in the Zeus repository. Short-term challenge results are typically available within about a week; long-term results take around three weeks due to the publication lag of reanalysis data. Validators use a commit-reveal protocol to prevent score manipulation. They compute area-weighted error metrics for each miner and submit weight vectors to Yuma Consensus.
Ground-Truth Scoring Context
A weather forecast can only be judged once the predicted period has actually occurred, so Zeus reveals scores only after ERA5 reanalysis data for the forecast window is published. The Zeus repository describes miners being graded against this authoritative external dataset rather than against any self-reported confidence, which is what makes the score a measure of real predictive accuracy. The area-weighted error metrics account for the spherical geometry of the global grid, so accuracy is measured fairly across latitudes instead of over-counting the more densely packed grid cells near the poles. A miner’s score therefore reflects how close its forecasts came to observed reality, weighted consistently across the Earth’s surface.
Consistency and Anti-Gaming Context
Zeus aggregates rankings over a rolling window of recent challenges rather than scoring any single submission, so a miner has to forecast accurately again and again rather than succeed once by chance. The Zeus repository describes a commit-reveal protocol that prevents a miner from altering a submission after seeing other entries or the eventual ground truth, keeping the comparison fair. The winner-take-most reward structure, in which the top-ranked miner captures the large majority of weight and runners-up receive logarithmically less, means that small but consistent gains in forecast accuracy translate into a much larger share of emissions.
On-Chain Identity
Zeus is registered at netuid 18 on Bittensor with 257 neurons, verifiable via
taostats.io/subnets/18. The subnet owner coldkey is
5DHwWLjtpwnZQUQKKXE2N5Gdy2N8PpqhgjLUuzgSB7yuGZkF. The codebase is at
Orpheus-AI/Zeus and the project site is
zeussubnet.com.
Relationship to Multiple Mechanisms
Zeus has validators score miner weather forecasts against ERA5 ground truth. The Glossary and Multiple Incentive Mechanisms Within Subnets docs note that validators must evaluate miners separately for each mechanism.
For readers, this article documents one subnet market. If that netuid runs more than one incentive mechanism, validator scores and weights should be read per mechanism rather than as one combined path.
Relationship to Yuma Consensus
Subnet 18 uses Yuma Consensus to convert the forecast-accuracy 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 Zeus’s context, the Zeus repository describes validators scoring miner weather forecast submissions against ERA5 reanalysis ground truth using area-weighted error metrics, aggregating rankings over a rolling window of recent challenges, and translating those ranked scores into on-chain weight vectors. 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 Zeus (SN18), that sequence changes how readers should interpret weather and climate forecast examples and ERA5-evaluated scoring outcomes.
In localnet, Zeus-compatible miners and validators can be developed and tested in an isolated environment. Localnet forecast evaluation results and emission outcomes do not represent production subnet performance.
On testnet, Zeus-compatible forecast submission and ERA5 validation workflows can be exercised in a shared, non-production network. Testnet evaluation scores and validator weights are separate from mainnet subnet state.
On mainnet, Zeus (SN18) is the live production subnet where miners produce global weather and climate forecasts and validators evaluate them against ERA5 reanalysis ground truth to determine real Bittensor emissions. The Zeus repository is the registered project repository for SN18 on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A weather forecast evaluation result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Miner and Validator Roles
Subnet 18 operates under the standard Bittensor two-role structure. Miners supply the subnet’s capability and validators evaluate those contributions and set weights. Reward distribution follows Yuma Consensus.
Reader Boundary
Subnet 18 Zeus should not be read as generic Bittensor subnet documentation, a live weather API with instant scores, or proof that one forecast defines long-term ranking. It names one subnet’s global environmental forecasting competition on netuid 18 (Understanding Subnets, Glossary: Netuid).
ERA5 Reanalysis Supplies Delayed Ground Truth
The Zeus repository scores miner forecasts only after ERA5 reanalysis data for the predicted window is published (Zeus repository).
Short- and long-horizon challenges therefore settle on different publication timelines rather than immediate submission time.
Area-Weighted Errors Account for Grid Geometry
Validators score forecasts with area-weighted error metrics so latitude differences across the global grid affect rankings appropriately (Zeus repository).
That weighting keeps polar and equatorial grid cells comparable within one evaluation pass.
Validator Weights Still Flow Through Yuma Consensus
Subnet 18 uses Yuma Consensus to convert validator weight submissions into emission shares each tempo (Yuma Consensus, Emission).