The Meta-Model (STMM)

The Sportstensor Meta-Model, or STMM, is the network's aggregate output. It is an ensemble architecture that takes the forecasts of every competing miner and synthesizes them into a single prediction that is more accurate than any individual model. It is the answer to a simple question: if hundreds of independent models are all predicting the same game, what does their combined intelligence believe?

// COLLECTIVE INTELLIGENCE

The founding principle is that collective intelligence beats any single model. No individual forecaster is right about everything — each has blind spots, biases, and regimes where it underperforms. But when many diverse, independently-built models are combined and weighted by demonstrated skill, their errors tend to cancel while their genuine signal reinforces. The ensemble is consistently sharper than its best individual member.

Because the subnet is model-agnostic, the diversity feeding the ensemble is real: neural networks, Monte Carlo simulations, quant strategies, autonomous agents, and human traders all contribute different views of the same event. That diversity is precisely what makes an ensemble powerful — a room of similar models adds little, but a room of genuinely different ones adds a great deal.

// PROVEN AGAINST THE MARKET

The STMM was featured in Grayscale and FS Insight's April 2025 report “Bittensor: The Internet of AI.” The report showed the Meta-Model significantly outperforming the NBA market over a three-month period — a public, third-party validation that the network's synthesized forecast carries a real, measurable edge over the crowd's price.

That matters because the benchmark is not a backtest or a paper score. The Meta-Model is measured against a live, liquid market of real participants putting real money behind their opinions. Beating that consistently is the hardest and most honest test a prediction system can face.

// WHO CONSUMES IT

The Meta-Model's intelligence is consumed by traders on Almanac and by a growing set of autonomous agents built on top of the network — among them BillyBets, Numinous, and Oddy — with a potential hedge fund on the horizon. Alongside sports, an in-development computer-vision initiative is building frame-level positional data from sports video to create novel proprietary datasets that let miners train even better models.

See also: The Subnet · Scoring & Incentives