Overview

Sportstensor is a decentralized AI network for sports prediction, operating as Subnet 41 on Bittensor. It runs a continuous, real-time competition where independent developers — miners — deploy predictive models, compete against real-world outcomes, and earn the network's native SN41 token based on measured performance. The network's aggregate output, the Sportstensor Meta-Model, synthesizes every competing model into a single forecast more accurate than any individual one, and Almanac monetizes that intelligence on live prediction markets.

// THE THESIS

Sports betting and prediction markets are dominated by specialized quant firms with enormous compute and capital. Sportstensor's premise is that sports are mathematical proving grounds — beneath the chaos there are persistent, exploitable patterns — and that the edge to find them should not be locked behind closed doors. The network democratizes algorithmic sports prediction: anyone can register a miner and put a model to work.

The system is model-agnostic. Neural nets, Monte Carlo simulations, quant strategies, autonomous agents, and humans trading by hand all compete on equal footing — only results matter. Models stay private and IP-protected; only predictions and trades are judged, never source code. The belief underneath it all is that collective intelligence beats any single model: no one forecaster is right about everything, but the ensemble is more accurate than its best member.

Crucially, every prediction is backed by a real trade on a real market — not a theoretical benchmark. That gives the network skin in the game and lets it reward being accurate before it's obvious, rather than converging on the answer once the crowd already has it. This is the difference between genuine intelligence and lucky streaks.

// THE FOUR PILLARS

Sportstensor is built from four interlocking parts. Each has its own page with the full detail.

The Subnet (SN41)

The network layer — the world's first decentralized competition network for sports prediction. Open and permissionless: anyone can register a miner or run a validator. Miners generate information signals by trading; validators sync metadata, ingest rolling trading data, run scoring, and set weights on-chain. Multiple independent validators make cheating difficult.

The Meta-Model (STMM)

An ensemble architecture that aggregates all competing miner models into a single synthesized forecast. In Grayscale's April 2025 report “Bittensor: The Internet of AI”, the Meta-Model was shown significantly outperforming the NBA market over a three-month period.

Scoring & Incentives

The incentive mechanism that decides who earns. Predictions are graded on closing edge, ROI, and risk-adjusted (Sortino) returns over a rolling 30-day window, with eligibility gates that filter out unprofitable noise. Two-phase optimization funds volume within a fixed token budget, then redistributes toward higher-ROI traders.

Almanac

The flagship product — an incentivized prediction market terminal that abstracts away all Bittensor complexity. Users trade sports outcomes and 5-minute crypto markets; trades route to Polymarket's order books in the background. Every trade is a scored prediction, and a 1% fee on winning trades buys back SN41.

// WHERE TO NEXT

Start with The Subnet to understand how the network is structured, then read The Meta-Model and Scoring & Incentives for how intelligence and rewards flow. For the product, see The Terminal and Tokenomics. Common questions are in the FAQ.