Score

What is Score?

Score, also known as sn44 or Score Vision, is a Bittensor subnet that applies decentralized machine-learning incentives to computer vision, initially by turning football and other video streams into structured machine-readable data such as player locations, ball tracking, pitch geometry, object detections, and event context.

The problem it addresses is not generic “AI compute” but the narrower and commercially meaningful bottleneck of video annotation: converting large volumes of raw footage into accurate labels quickly enough and cheaply enough to be useful for sports analytics, safety monitoring, retail operations, logistics, and other camera-heavy environments.

Its claimed moat is the combination of Bittensor’s miner-validator incentive market with lightweight validation methods, including frame filtering, pitch/keypoint checks, homography-style geometry tests, and CLIP-based semantic verification, which are designed to avoid the cost of re-running full vision inference on every submitted frame.

The project’s own GitHub repository describes Score Vision as a decentralized computer-vision framework focused first on Game State Recognition in football, while the current Bittensor subnet page characterizes sn44 as a framework in which miners process video locally and validators score results through hybrid visual and geometric checks. (github.com)

Score’s market position is best understood as a specialized Bittensor application subnet, not as a base-layer blockchain or broad smart-contract platform.

As of late June 2026, public market screens placed Score in the mid-cap range among liquid Bittensor subnet tokens rather than among the largest crypto networks; CoinGecko’s recent crawl showed Score ranked around the low-500s by crypto market capitalization, while Bittensor.ai’s live subnet view showed a subnet with a full 256/256 neuron set, nine validators, several thousand holders visible on Taostats, and roughly 131,000 TAO-equivalent TVL in the subnet pool snapshot. Those figures should be treated as point-in-time market and staking indicators, not evidence of durable end-user demand. More analytically, Score’s scale is still small relative to centralized computer-vision vendors and sports-data incumbents, but it is comparatively differentiated inside Bittensor because it targets a measurable external output—vision models and video-derived labels—rather than a purely speculative emissions game. (coingecko.com)

Who Founded Score and When?

Score appears to have emerged publicly in 2024, during the post-ChatGPT expansion of AI-infrastructure narratives and the early Bittensor subnet-token cycle.

The project’s company presence lists Score - Subnet 44 as founded in 2024 and headquartered in New York City, while the Bittensor subnet record shows sn44 registered on-chain in September 2024. Founder attribution varies slightly across public materials, but the most consistent names are Maxime Sebti, Tim Kalic, and Nigel Grant; SIRE documentation identifies Maxime Sebti as co-founder and CEO of Score Technologies, Tim Kalic as co-founder and CTO, and Nigel Grant as co-founder and chief revenue officer, while LinkedIn references Tim Kalic as co-founder and CTO of Score - Subnet 44 and Manako Labs. The operating entity is often described as Score Technologies or Vision Research Foundation-linked, with Manako Labs later becoming a visible commercial interface built on top of the subnet. (linkedin.com)

The project’s narrative has shifted materially since launch. Early community material around Score was closer to sports prediction, sports analytics, and football-community onboarding, whereas the current positioning is broader: an “open, permissionless computer vision layer” that can train and evaluate small, task-specific vision models for real-world camera networks.

The football Game State Recognition thesis remains important because sports footage offers dense, high-value labeled data and a clear commercial market, but the more recent Manako framing moves Score toward enterprise physical-AI use cases such as restricted-zone alerts, fuel-station object detection, vehicle/person detection, and edge-deployed operational monitoring. This evolution is strategically rational, because pure sports analytics is a niche market with entrenched incumbents, while enterprise camera intelligence is larger, but it also raises execution risk: Score must prove it can generalize beyond football without losing the validation rigor that made the original subnet design coherent. (kucoin.com)

How Does the Score Network Work?

Score does not operate an independent proof-of-work, proof-of-stake, or DAG blockchain. It is an application-specific subnet running on Bittensor’s Subtensor L1, where the relevant “consensus” for Score is Bittensor’s stake-weighted Yuma Consensus process rather than a standalone block-production mechanism. In Bittensor, subnets are incentive marketplaces: miners perform a defined AI task, validators evaluate the quality of that work, and Yuma Consensus converts validator weight submissions into miner and validator emissions.

The Bittensor documentation states that Yuma Consensus runs on-chain within Subtensor and computes miner and validator emissions from validators’ rankings of miner performance, with stake-weighted clipping intended to reduce collusive or unreliable scoring. For sn44, that means the security model is partly inherited from Bittensor’s chain and partly dependent on whether Score’s validators can reliably distinguish high-quality computer-vision output from low-quality or adversarial submissions. (docs.learnbittensor.org)

Technically, Score’s architecture is a three-role system: miners receive video or image tasks and run object detection, tracking, or specialized model inference locally; validators sample and score miner outputs; and the subnet owner maintains task design, incentive parameters, and overall network health.

The distinctive feature is the validation approach. Instead of validating every frame with expensive full-model inference, Score uses filtered frames, semantic checks, keypoint and pitch-geometry plausibility, reprojection error, and GS-HOTA-style detection-association metrics to approximate quality efficiently.

Earlier Score materials emphasized football clips, player-ball detection, pitch-line extraction, and 30-second match segments; newer materials emphasize model distillation and lightweight, edge-deployable vision skills. This is technically plausible, but it creates a central tension: the more Score expands into arbitrary enterprise vision tasks, the harder it becomes to maintain a single robust validation regime, and the more the subnet depends on careful benchmark design rather than simply adding more miners. (github.com)

What Are the Tokenomics of sn44?

sn44 is an alpha token under Bittensor’s Dynamic TAO model, so its supply and value mechanics differ from a conventional ERC-20 with a fixed allocation table. Bittensor’s Dynamic TAO FAQ states that each subnet alpha token has a 21 million hard cap and follows a halving schedule, while the emissions documentation explains that subnet-specific alpha tokens are emitted to miners, validators, stakers, and subnet creators. As of late June 2026, third-party market pages indicated roughly 4–5 million SN44 in circulating supply and a market capitalization in the high-$30 million to low-$40 million range, while the user-provided asset snapshot placed market capitalization around $42.4 million and the token in the high-single-digit dollar range. Structurally, sn44 is inflationary until emissions decay through halvings and supply approaches the cap; it is not primarily a burn-token model, although Bittensor registration costs and protocol-level mechanisms can affect TAO/alpha flows around subnet participation. docs.learnbittensor.org

Value accrual comes from staking demand, miner-validator economics, and the market’s assessment of whether the subnet produces valuable computer-vision outputs. In Dynamic TAO, a user staking into a mining subnet effectively exchanges TAO for that subnet’s alpha and stakes that alpha to a validator; exit value then depends on the alpha-to-TAO pool ratio at unstaking. Bittensor’s June 2026 emissions documentation is important because it says the network had reverted to a price-based model for distributing TAO emissions across subnets, after a November 2025 to June 2026 flow-based period, which means subnet token prices and moving averages again influence emission share.

For Score specifically, Bittensor.ai’s late-June snapshot showed an 18% owner cut and emissions split among miners, validators/stakers, and the owner, with very high displayed staking APY that should be interpreted as a volatile emissions output rather than a stable yield. In economic terms, sn44 holders are underwriting a reflexive system: useful models and external demand can justify stake inflows and emissions, but emissions without fee-paying demand can dilute holders and reward short-term capital rotation rather than durable network utility. (docs.learnbittensor.org)

Who Is Using Score?

The key distinction is between token activity and product usage. Score’s on-chain trading volume, holders, validator count, and staking TVL show that the asset has market participation, but those metrics do not prove that enterprises or sports teams are paying for vision outputs.

Actual utility is better inferred from task activity, model benchmarks, miner competitions, and commercial applications built on the subnet. Score’s public materials identify sports analytics, broadcasting, betting, scouting, and coaching as initial target markets, while more recent public communications describe broader computer-vision tracks such as person detection, vehicle detection, fire detection, and fuel-station monitoring.

As of mid-2026, the most credible use pattern is not retail users interacting directly with sn44, but builders using the subnet as a decentralized model-discovery and model-distillation backend. (github.com)

The most concrete enterprise-facing adoption signal is Manako Labs. In April 2026, Manako announced an alliance with PwC France and Maghreb, saying PwC France would lean on Manako’s Business Operations World Model powered by Score - Subnet 44 to help organizations turn existing camera networks into operational intelligence systems. In June 2026, a CryptoBriefing item syndicated by KuCoin reported that Manako had launched a vision AI agent platform powered by Bittensor’s Score Subnet 44, with a no-code interface, CPU-runnable models, edge processing, Slack alerts, and a stated $1 million TaoWeave investment for North American expansion. These are meaningful commercial signals, but they are not the same as audited revenue, customer retention, or enterprise-scale deployment metrics. A skeptical reading is that Score has promising distribution through Manako and PwC-adjacent advisory channels, but still needs to disclose stronger evidence of repeat customers, paid workloads, and throughput measured in processed camera-hours or accepted model tasks. (manako.ai)

What Are the Risks and Challenges for Score?

Score’s regulatory exposure is indirect but real. There does not appear to be a known active regulator lawsuit specifically against Score or sn44 in the public sources reviewed, but sn44 inherits the broader uncertainty around TAO, Bittensor subnet tokens, staking, and emissions-driven digital assets. Grayscale’s filed Bittensor Trust S-1 explicitly discusses the risk that TAO could be argued to be a security and notes that the SEC or a court could take a contrary view, even where the sponsor views TAO as not being a security. That matters for sn44 because alpha tokens are even more tightly linked to subnet creator activity, emissions design, staking flows, and expectations around productive managerial effort. Centralization is the second major risk. The Bittensor.ai page showed only nine validators on sn44 in its late-June snapshot, an 18% owner cut, disabled commit-reveal and liquid-alpha settings, and a health label that described the subnet as abandoned while also showing no GitHub commits in the preceding 30 days and a last commit roughly 200 days earlier. Some of those labels may lag off-chain development, but institutional investors should treat validator concentration, owner discretion, stale repositories, and opaque task governance as material due-diligence items. sec.gov

Competitive risk is also substantial. In sports analytics, Score competes economically with incumbent data and video-analysis providers such as Opta-style sports-data vendors, club analytics stacks, broadcast-tracking systems, and specialized computer-vision providers that do not need crypto incentives. In enterprise vision, it competes with cloud AI platforms, edge-AI vendors, Roboflow-like tooling, open-source models, and proprietary vertical solutions embedded into security, retail, logistics, and industrial software. The decentralized subnet model can be a cost and talent-discovery advantage if it reliably sources better models, but it can also be slower to productize than a centralized vendor with direct customer feedback loops, service-level agreements, procurement teams, and compliance controls. The token adds another threat: if emissions are more attractive than external revenue, miners and validators may optimize for reward mechanics rather than customer outcomes, creating a gap between subnet activity and economically useful output. medium.com

What Is the Future Outlook for Score?

Score’s outlook depends less on price performance and more on whether it can convert a credible technical niche into repeatable commercial infrastructure.

The verified roadmap in the public GitHub materials laid out a 2025 sequence from Game State Recognition and VLM-based validation to mainnet deployment, human-in-the-loop validation, dashboards, action spotting, event captioning, integration APIs, additional sports, developer tools, and cross-domain applications.

By mid-2026, the public narrative had advanced toward Manako-powered enterprise camera intelligence and small task-specific model distillation, while Bittensor itself had undergone important tokenomics changes, including the June 2026 return to price-based emissions.

The most important milestones from here are therefore practical rather than promotional: refreshed open-source development, clearer validator and miner telemetry, audited model benchmarks, public API documentation, evidence of paid workloads, and a robust validation framework for non-football tasks. (github.com)

The structural hurdle is that Score must prove the subnet is more than an emissions-subsidized model competition.

If Manako and similar applications can consistently route real enterprise vision problems into sn44, benchmark miner outputs, deploy compact models at the edge, and show cost or accuracy advantages over centralized tooling, then Score has a defensible role as a Bittensor-native computer-vision labor market. If not, the asset risks being valued mainly as a leveraged bet on Bittensor subnet speculation, with limited separation between token liquidity and actual product-market fit. No price forecast is warranted; the relevant question is whether sn44 can sustain high-quality validation, decentralize control, and turn camera data into externally demanded infrastructure before centralized vision platforms close the cost-efficiency gap.