The Subnet (SN41)

Sportstensor is Subnet 41 on Bittensor — the world's first decentralized competition network for sports prediction. It uses Bittensor as its base layer for incentive distribution, consensus, and peer-to-peer machine-learning communication, so there is no bespoke blockchain to maintain. The subnet is open and permissionless: anyone can register a miner UID or run a validator and take part in the competition.

// MINERS

Miners generate information signals by trading. The subnet is model-agnostic — a miner can be a neural network, a simulation, a quant strategy, an autonomous agent, or a person trading manually. Only results are judged; the model itself stays private and IP-protected. High-signal miners earn the largest share of daily SN41 emissions, while unprofitable activity earns nothing.

Getting started as a miner requires an Almanac account linked to Polymarket, a registered Bittensor coldkey, and Python 3.10+ (CPU-only — no GPU needed). Registering a miner UID burns TAO, which keeps the network honest by making sybil registration costly.

// VALIDATORS

Validators are the network's referees. Each epoch they sync miner metadata from chain — wallets, proxy addresses, and UIDs — ingest the rolling trading data from the Almanac backend, run the two-phase scoring process, and set weights on-chain. Scoring runs hourly in the background against 24-hour epochs, always scoring the previous epoch while updating current weights.

Because multiple independent validators score the same data and set weights independently, no single actor can manipulate outcomes — which makes cheating difficult. Validators support Weights & Biases logging and ship with pm2 auto-update scripts for hands-off operation.

// WHY DECENTRALIZED

A centralized prediction service is only as good as its one model and only as trustworthy as its one operator. A decentralized subnet flips both problems: it invites an open field of competitors to out-predict each other, and it distributes the job of judging them across many validators. The result is a network that gets sharper as more intelligence joins, and whose scores no single participant can fake.