Subnet Scoring Model

Subnet Scoring Model
Area	Subnet incentives
Evaluates	Miner responses
Used by	Validators
Separate from	Consensus output

A subnet scoring model is the evaluation method validators use when judging miner responses inside a Bittensor subnet (Glossary: Subnet Scoring Model, Understanding Incentive Mechanisms).

The term belongs to incentive-mechanism vocabulary. It names how responses become evaluation scores, so validators can turn returned work into signals used later in the incentive process.

Evaluation Method

The scoring model explains how returned work is judged. The subnet task describes what miners are asked to produce, while the scoring model describes how validators evaluate the responses they receive (Understanding Incentive Mechanisms, Glossary: Subnet Task, Glossary: Subnet Scoring Model).

Task vocabulary and scoring vocabulary stay separate. The task asks for work; the scoring model evaluates returned work.

The scoring model is therefore the evaluation method, not the whole incentive mechanism. It turns responses into scores, while other parts of the system handle interaction, aggregation, emissions, and role-specific outcomes.

Task and Protocol

Subnet task, subnet protocol, and subnet scoring model describe adjacent pieces of the same flow. The task names the work target, the protocol describes the request-and-response pattern, and the scoring model describes how returned responses are judged (Glossary: Subnet Protocol, Glossary: Subnet Scoring Model).

That distinction keeps work design, interaction rules, and evaluation criteria from collapsing into one phrase. A subnet can change how work is requested or represented without making every protocol detail part of the scoring model.

The protocol creates the interaction surface. The scoring model explains the evaluation method applied to that surface.

Scores Before Weights

A scoring model sits earlier in the evaluation path than final weight signals. Weight vectors are structured signals formed after validator evaluation and sent onward for consensus processing (Glossary: Weight Vector, Glossary: Subnet Scoring Model).

This distinction keeps scoring formulas and weight vocabulary separate. The scoring model produces evaluation scores; weight vocabulary describes the signals that later carry evaluation into consensus.

A score is an assessment of returned work. A weight vector is the structured validator signal that can carry those assessments into the downstream consensus process. Keeping those terms separate helps explain why a scoring method is earlier than final incentive outcomes.

That order gives the term its narrow scope. The scoring model belongs to the evaluation step, while weight vectors and Yuma Consensus describe later signal and aggregation steps (Glossary: Weight Vector, Yuma Consensus).

Miner and Validator Roles

Miners produce responses for the subnet, and validators evaluate those responses under subnet standards. Miners and validators are separate subnet roles (Understanding Subnets, Glossary: Subnet Miner, Glossary: Subnet Validator, Glossary: Subnet Scoring Model).

The scoring model is the method validators apply to that work. It links miner responses to validator evaluation without replacing either role.

Role language stays clearest when those parts remain distinct. Miner describes the party producing work, validator describes the party evaluating work, and subnet scoring model describes the method used to judge returned work.

Mechanism Scope

When a subnet uses multiple incentive mechanisms, each mechanism can have its own scoring method. The multiple-mechanism documentation describes separate mechanisms inside one subnet (Multiple Incentive Mechanisms Within Subnets, Glossary: Subnet Scoring Model).

A scoring standard for one mechanism belongs to that mechanism. It does not automatically describe every task, evaluation path, or incentive pathway inside the subnet.

Mechanism labels therefore matter when comparing scoring examples.

This matters because the same subnet can contain more than one pathway for work and evaluation. Concrete scoring examples belong with the subnet, mechanism, network, and source that produced them.

Consensus Path

Yuma Consensus is downstream of validator evaluation. A scoring model can shape the evaluation a validator performs, while consensus aggregates validator signals into incentive and dividend outcomes (Yuma Consensus, Understanding Incentive Mechanisms).

This keeps evaluation methods separate from final consensus outputs. Scoring is part of how work is judged; consensus is the process that aggregates validator signals.

The distinction also keeps payout language precise. A scoring model can influence the evaluation that enters validator signals, but emissions, incentives, and dividends depend on downstream mechanism behavior.

Network Reading

The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. The subnet scoring model concept applies across the Bittensor lifecycle: evaluation methods can be developed in localnet for isolated testing, exercised on testnet in a shared non-production environment, and applied on mainnet for live emission-relevant scoring.

Bittensor network material separates mainnet, testnet, and localnet. Scoring model examples or evaluation outcomes from one environment do not automatically describe production subnet performance in another environment (Bittensor Networks).

Localnet examples are isolated development examples. Testnet examples are shared non-production examples. Mainnet scoring interpretation concerns live subnet behavior, production validator evaluation, and the incentive mechanism actually running on that subnet.

Development Stage Context

The localnet, testnet, and mainnet environments the Bittensor Networks reference separates matter to a scoring model on the read side: the documented scoring method is stable across them, while any observed score belongs to the network that produced it, so a localnet or testnet result is not evidence of a miner’s mainnet standing.

In localnet, subnet Scoring Model can be exercised in an isolated development environment, where the surrounding chain state reflects local configuration rather than production history.

On testnet, subnet Scoring Model can be observed in a shared, non-production network whose state is kept separate from mainnet.

On mainnet, subnet Scoring Model applies on the live, production Bittensor network, where the surrounding state is real and persistent.

The Bittensor Networks reference separates mainnet, testnet, and localnet. A subnet Scoring Model example from one environment should not be read as representing another environment.

Relationship to Yuma Consensus

Subnet Scoring Model and Yuma Consensus describe related parts of Bittensor’s incentive system. Yuma Consensus is the on-chain process that aggregates validator weight signals within a subnet into miner incentives and validator dividends, applying consensus clipping, bonding, and emission calculation (Yuma Consensus).

For readers, subnet scoring model names a specific part of that incentive picture, while Yuma Consensus names the consensus process that turns validator weights into the resulting incentives and dividends.

Reader Boundary

Subnet scoring model should not be read as benchmark selection, deployment status, or a full subnet design. It names concept vocabulary for how miner responses are evaluated inside a subnet incentive mechanism (Understanding Incentive Mechanisms, Understanding Subnets).

Subnet task names the work target. Subnet protocol names the validator-miner exchange. Subnet scoring model names how returned work is evaluated before signals move toward weight vectors and Yuma Consensus (Glossary: Weight Vector, Yuma Consensus).

Mechanisms Require Task, Protocol, and Scoring Together

Official Understanding Incentive Mechanisms documentation states that each mechanism must supply a protocol for validator-miner exchange, a task definition, and a scoring method validators can apply to returned work. The scoring model is therefore one required design piece rather than an optional add-on to task or protocol vocabulary.

Subnet creators define all three before evaluation can produce meaningful validator signals for a subnet market.

Scoring Precedes Weight Vector Submission

The Glossary: Weight Vector names the structured miner scores a validator submits after evaluation. A subnet scoring model explains how returned work becomes those scores; the weight vector carries the scores into the subnet-level weight matrix that Yuma Consensus reads (Yuma Consensus).

That ordering keeps method and signal separate. Scoring names the evaluation rule; weight vocabulary names the submitted consensus input built from that evaluation.

Clipping Filters Scores After They Enter Consensus

Yuma Consensus clipping compares submitted weights against a stake-weighted benchmark and trims values above that level. Scoring models therefore sit upstream of filtering: they shape validator judgments, while clipping decides how much of each judgment can affect final rank and incentives.

A high local score from a scoring model does not by itself guarantee unchanged influence after consensus processing.