Subnet 46: Zipcode
Zipcode is Bittensor Subnet 46. Public RESI materials describe it as a real-estate price-prediction subnet where miners submit residential property models and validators evaluate those models against fresh sales data.
What Zipcode Provides
The RESI README frames the subnet around accurate United States residential property price prediction. Miners build models that estimate property sale prices, while validators compare those predictions with ground-truth sales data.
At the subnet level, Zipcode turns model generalization into a miner-scoring signal. The README describes a temporal separation between model commitment and evaluation: models are committed before the sales data used for evaluation is available, so a model has to generalize to newly listed and sold properties rather than memorize a static dataset.
Submission and Evaluation Flow
The miner guide describes miner participation as training an ONNX model, testing it locally, uploading it to HuggingFace, and submitting a commitment to the Bittensor chain. Validators later download the model, verify that it matches the commitment, and score it based on prediction accuracy.
The validator guide describes validator responsibility in three reader-facing steps: download miner ONNX models, evaluate them against daily validation data, and submit scores to the Bittensor blockchain. This makes the validator role a model-evaluation and weight-setting layer rather than a data-provider role.
That flow is important because the subnet rewards a committed model artifact rather than a single manual prediction. The model can be independently fetched, run in validation, and compared with later real-estate outcomes.
Miner and Validator Roles
Miners train and submit property price-prediction models. The miner guide describes ONNX as the model format and explains that models use property features, with optional image inputs, to produce predicted prices. The practical miner contribution is therefore a reusable forecasting model, not a manual price guess for each property.
Validators fetch submitted models, run inference, calculate prediction scores, rank models, and set weights on chain. The README describes scoring with Mean Absolute Percentage Error, so lower prediction error produces a stronger score. It also describes a winner-selection design where models within a threshold of the best score are compared by commit timing, which is meant to preserve an advantage for the model that first reached the stronger performance tier.
Evaluation Context
Zipcode’s evaluation context is real-estate generalization. A model that performs well only on old or memorized examples should not be competitive when the validation set comes from properties that were listed and sold after the model was committed. The unseen-data gap is why the subnet emphasizes model commitment, validation timing, and ground-truth sales outcomes.
The reward design also makes model duplication less attractive. The README says detected duplicate models receive no reward, while non-winning valid models share a much smaller portion of emissions. In practice, Subnet 46 tries to incentivize earlier, genuinely better property-price models rather than many near-identical copies of a current leader.
Relationship to Yuma Consensus
Subnet 46 uses Yuma Consensus to convert the property-price model scores 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 Zipcode’s context, validators evaluate committed ONNX models against real-estate sales data and translate prediction accuracy into on-chain weights. The Emission documentation describes how those consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
On-Chain Identity
Live SN46 subnet data is available on TaoStats. The source-backed real-estate, model-submission, and evaluation details in this article come from public RESI materials rather than from live identity fields.
Relationship to Yuma Consensus
Subnet 46 uses Yuma Consensus to convert the prediction-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 Zipcode’s context, validators download committed miner ONNX models, evaluate them against daily validation data using Mean Absolute Percentage Error, rank models by prediction accuracy with commit-timing tiebreakers, and translate those ranked scores into weight vectors for the subnet. Because those weight vectors come from accuracy rankings, a miner’s emission share follows from how its model performs against the validation data rather than from having committed a model alone (Emission).
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Zipcode (SN46), that sequence changes how readers should interpret real-estate price prediction examples and model evaluation outcomes.
In localnet, Zipcode-compatible miners and validators can be developed and tested in an isolated environment. Localnet model evaluation scores and emission outcomes do not represent production subnet performance.
On testnet, Zipcode-compatible property price prediction models can be exercised in a shared, non-production network. Testnet evaluations and validator scores are separate from mainnet subnet state.
On mainnet, Zipcode (SN46) is the live production subnet where miners submit residential property price-prediction models and validators score those models against real sales data to determine real Bittensor emissions. The RESI-models repository describes the mechanism that applies on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A price prediction result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Reader Boundary
Subnet 46 Zipcode should not be read as generic Bittensor subnet documentation, a real-estate brokerage service, or proof that one manual price guess counts as a miner submission. It names one subnet’s residential property price-prediction model competition on netuid 46 (Understanding Subnets, Glossary: Netuid).
ONNX Commitments Bind Reusable Model Artifacts
The miner guide describes miners training an ONNX model, uploading it to HuggingFace, and submitting a commitment on the Bittensor chain for validators to verify (miner guide).
Validators therefore score committed model artifacts rather than one-off predictions.
MAPE Ranks Models Against Daily Validation Data
The validator guide describes downloading miner ONNX models, evaluating them against daily validation data, and ranking models by prediction accuracy (validator guide).
Mean Absolute Percentage Error is the comparison metric named in the project materials.
Validator Weights Still Flow Through Yuma Consensus
Subnet 46 uses Yuma Consensus to convert validator weight submissions into emission shares each tempo (Yuma Consensus, Emission).