Your Simple Guide to Score (SN44)

Analyzing a single football match manually takes hundreds of hours of human work. Annotating who is on the pitch, where they move, and what happens costs between $10 and $55 per minute of footage, according to project estimates. For broadcasters, betting operators, and clubs that want to analyze thousands of games, those numbers make comprehensive video intelligence practically unaffordable. Score changes this by building a decentralized computer vision network on Bittensor, where anyone can contribute computing power to process video and earn rewards for doing it well.

What is Score?

Score is a computer vision subnet running as Subnet 44 (SN44) on Bittensor. Computer vision is a field of AI that teaches machines to understand what they see in images and video. Instead of a human watching footage and logging what happens, a computer vision model does that work automatically. Score builds and coordinates those models across a decentralized network. Score Technologies Inc., the company behind the project, describes the mission as making every camera intelligent.

The first application Score focused on is Game State Recognition in football. This means the network watches match footage and tracks every player, the referee, and the ball across each frame of the game in real time. It maps who is where and how everyone is moving, turning raw video into data that coaches, scouts, and broadcasters can use. Football is a deliberate entry point.

The sport sits within a $600 billion global industry, with roughly $50 billion in betting and $30 billion in sports data services, based on figures referenced by the project. Moreover, the analysis tools that exist today are expensive enough that only the largest clubs can afford them. Score targets exactly that access gap.

The framework is designed to work beyond football as well. Basketball, tennis, security cameras, and retail environments all share the same core need: understanding what cameras see, quickly and at low cost. Consequently, Score treats football as a proving ground rather than an endpoint. In late 2025, the team signed a partnership with Two-a-Day to apply its Vision AI to production operations.

How Score works?

Score coordinates three roles to process video across the network. First, miners receive video footage and run it through AI models that detect and track objects frame by frame. They produce a continuous output describing what is happening on the pitch at every moment. Second, validators check the quality of that output without watching the footage themselves. Third, the Subnet Owner monitors overall network health and adjusts reward parameters based on performance.

The validation step is where Score’s core innovation sits. Checking whether a miner correctly identified every player in 90 minutes of footage would normally require watching all of it. Score solves this with a two-stage lightweight process. In the first stage, it selects a sample of frames and checks whether player positions and pitch landmarks are consistent and plausible. In the second stage, an AI model verifies that the objects in those frames are correctly labeled, distinguishing players, the ball, referees, and goalkeepers. Both checks produce a combined quality score for each miner. As a result, validators assess accuracy at significantly lower cost than traditional methods.

Score measures miner performance using a metric called GS-HOTA (Game State Higher Order Tracking Accuracy). It captures both detection quality and tracking consistency across frames. Together, these two dimensions give a clear picture of how useful a miner’s output actually is. Rewards reflect quality, consistency, and response speed. Validators earn based on how closely their assessments align with the network consensus. High performers earn more, and low-quality work leads to reduced rewards.

Who is behind it?

Score Technologies Inc. built and operates Score Vision. Maxime Sebti co-founded the company and serves as CEO. Dr. Peter Cotton co-founded the company and leads the research team. Together, they built Score around the conviction that professional-grade computer vision should be accessible to any organization that uses cameras, not only those with the largest technology budgets.

All code is published on GitHub under an MIT license, so anyone can read, audit, or build on the codebase. The team also published a research paper titled “Score Vision: Enabling Complex Computer Vision Through Lightweight Validation – A Game State Recognition Framework for Live Football.” Score Vision launched on the Bittensor mainnet under netuid 44 in January 2025.

The project maintains active communities on Discord, X: @webuildscore, and LinkedIn. A live dashboard tracking network activity is available at console.scorevision.io.

Why it is valuable?

Score addresses a problem of cost and speed simultaneously. Today, detailed video analysis of football is accessible only to elite clubs. Analyzing hundreds of thousands of matches across global leagues requires a budget and a team that most organizations simply do not have. Even where money is available, human annotation is too slow for live use cases such as in-play betting or real-time broadcast statistics.

Score distributes that workload across a global network of miners and validates the results automatically. According to the project’s own estimates, this reduces annotation costs by a factor of 10 to 100. Furthermore, according to CEO Maxime Sebti, as reported by Business Cloud, Score processes football video 240 times faster than current computer vision methods. As a result, semi-professional leagues, smaller broadcasters, and regional betting operators can access the same quality of data that was previously available only to the top tier of the sport.

Real-world adoption confirms this direction. In November 2025, Score announced a partnership with Reading FC, making the club the first professional football team in England to appoint a dedicated head of AI. The collaboration covers performance analysis, recruitment, and tactical planning. Former Reading manager Brian McDermott also contributes to the project. Reading is precisely the kind of club that previously could not afford this level of analysis.

Additionally, Score opens a new earning opportunity for anyone with a capable GPU anywhere in the world. Miners join the network, contribute video processing work, and earn rewards based purely on the quality of their output. Performance determines earnings, not location or affiliation.

The future of Score

The Score roadmap builds toward a more complete understanding of what happens in a match. Beyond tracking positions, the next steps include recognizing specific events such as goals, fouls, and passes, and generating automatic descriptions of match moments in natural language. Advanced player tracking and integration APIs for external platforms are also in development.

Beyond football, Score plans to extend the same framework to basketball, tennis, security surveillance, and retail analytics. These expansions address the same fundamental challenge: making sense of what cameras capture at scale and at low cost.

On the research side, the team is working toward an open-source vision-language model built specifically for video understanding. If realized, this would give the wider developer community a foundation to build on, multiplying the reach of the network beyond what the core team can build alone.

Sources:
https://www.wearescore.com
https://console.scorevision.io
https://github.com/score-technologies/score-vision
https://businesscloud.co.uk/news/reading-fc-to-score-with-ai-deal/
https://rdg.today/reading-fc-become-first-professional-club-to-appoint-head-of-ai/
https://www.skysports.com/football/news/11095/13483738/readings-head-of-ai-explains-ambition-to-be-pioneers-as-league-one-club-looks-for-an-edge-to-reach-the-premier-league