Network Design

Heterogeneous nodes, dual-engine proof of contribution, mining pool rewards, and data security.

3.1 Heterogeneous Nodes and Task-Splitting Architecture

The distributed computing network of Corex Token is built on globally distributed heterogeneous nodes, covering GPUs, CPUs, FPGAs, and edge computing devices, with the aim of breaking the monopoly and resource waste of centralized compute platforms. Through modular design, the network splits AI training tasks into executable fragments, enabling multi-node collaboration and ensuring that tasks are verifiable, traceable, and fairly rewarded.

The task-splitting module includes three parts:

• Task slicer: large AI model training tasks are broken down into tiny computation fragments, each containing verifiable gradient calculations, inference steps, and data processing indicators.
• Distributed execution layer: tasks are assigned to different nodes for execution and are intelligently matched based on node compute power, online duration, device type, and reputation score, ensuring that high-performance nodes are prioritized for high-compute tasks while small and medium-sized devices can handle lightweight computation and inference tasks.
• On-chain verification and reward calculation: after each task is executed, a Proof-of-Compute or Proof-of-Model-Training workload proof is generated, and rewards are automatically calculated and distributed to nodes through smart contracts.

The network is compatible with multiple hardware types, and nodes can connect without requiring professional mining farms. Automated node detection tools can evaluate compute performance, stability, and network connectivity to ensure that each node contributes genuine and reliable computing capability. At the same time, distributed task splitting not only improves the network's computational efficiency, but also ensures fairness in node participation, allowing global compute resources to be fully utilized and enabling high-concurrency, multi-node collaborative AI training.

In terms of architectural design, the Corex Token network emphasizes task-splitting granularity, node diversity, and load balancing. Task-splitting granularity determines the network's parallel processing capability: the finer the granularity, the more tasks each node can handle simultaneously, thereby improving training efficiency. Node diversity ensures that everything from high-performance GPUs to idle household CPUs can participate. The load-balancing algorithm dynamically assigns tasks based on node tiers, contribution history, and real-time load, avoiding node overload or task execution delay. Through these mechanisms, Corex Token builds an efficient, stable, and scalable distributed computing network that provides reliable infrastructure for decentralized AI training.

3.2 Dual-Engine Proof-of-Contribution Mechanism

To ensure that compute contributions are genuine and trustworthy while also incentivizing long-term node participation, Corex Token adopts a dual-engine proof-of-contribution mechanism that combines Compute-PoC, or Proof of Compute, and Model-PoT, or Proof of Model Training, supplemented by Stake-Boost to enhance node reward weights.

• Compute-PoC: base proof of compute contribution. Nodes complete foundational computing tasks through GPUs, CPUs, or edge devices and submit verifiable on-chain proof of work. Rewards are calculated based on the node's actual compute power, online time, task execution quality, and historical reputation, ensuring that invalid computation or repeated tasks do not generate rewards.
• Model-PoT: proof of AI model training contribution. Nodes carry out AI work such as gradient computation, model fragment training, and inference tasks. Smart contracts verify training quality and contribution size on-chain. Rewards are linked to the accuracy, efficiency, and stability with which nodes complete training tasks.
• Stake-Boost: stake enhancement mechanism. Nodes obtain task acceptance rights and verification weight by staking CXT. Staking itself does not directly mint tokens, but it improves a node's task allocation priority and reward ratio, granting additional incentives to long-term contributors while laying the foundation for future DAO governance.

This dual-engine mechanism not only guarantees the authenticity of compute output and AI training contribution, but also encourages long-term node participation in network construction, realizing the closed-loop economic logic that contribution means income. Through the combination of Compute-PoC, Model-PoT, and Stake-Boost, the Corex Token network establishes a trustworthy, quantifiable, and incentivizable global compute ecosystem and provides high-quality decentralized infrastructure for AI model training.

3.3 Core Mining Pool and Reward Strategy

The Corex Token network adopts a single-mining-pool structure to ensure that the token release process is transparent, decentralized, and free from human intervention. All 100,000,000 CXT tokens are minted at one time before the mainnet launch and locked into the mining pool smart contract. The mining pool has completely relinquished its permissions and can neither mint additional tokens, stop issuance, nor adjust release rules, thereby ensuring fairness and immutability in the network's token distribution.

The mining pool reward mechanism includes two main paths:

• Output from compute contribution: nodes complete computing tasks through Compute-PoC, and the mining pool calculates rewards based on node contribution values.
• Rewards for staking-based task acceptance: nodes stake CXT and obtain rights to accept advanced tasks and perform verification, then receive rewards by completing Model-PoT training tasks or verification tasks.

The mining pool contract comprehensively calculates node rewards dynamically based on compute performance, task completion quality, historical reputation, and staking weight. All rewards are executed automatically through on-chain smart contracts without manual intervention, ensuring transparency and traceability throughout the entire process. The mining pool strategy also includes a dynamic adjustment mechanism that automatically optimizes the reward distribution ratio based on the scale of network compute power and task demand, tightly binding network incentives to contributions and ensuring long-term sustainability.

In addition, the mining pool is linked with the node tier system, task priority rights, and the ecosystem incentive fund, allowing nodes not only to obtain immediate rewards but also to receive additional incentives in network governance, priority task acceptance, and long-term returns, thereby building a healthy and efficient decentralized compute economy.

3.4 Data Privacy and Security Protection

In a decentralized AI training network, data security and privacy protection are of critical importance. The Corex Token network ensures that node contributions are genuine and reliable while protecting task data privacy through a four-layer security strategy:

• Node isolation and reputation system: each node establishes an on-chain reputation profile that records successful tasks, failure rate, reward claims, staking status, and abnormal behavior. The higher the node's reputation, the higher its task allocation priority. Nodes that violate rules or behave maliciously will be automatically penalized or removed.
• Compute authenticity protection: through GPU fingerprint recognition, behavioral analysis, task replay verification, and gradient consistency checks, the network prevents fake compute power, simulated compute power, or repeated task submissions.
• Privacy protection: training data is executed only locally on nodes and is neither uploaded nor broadcast. Only parameter hashes, contribution proofs, and verification results are stored on-chain, ensuring the privacy of user data and model weights.
• Contract and communication security: the mining pool and task contracts are audited by professional third parties, and all task data transmission uses encryption and sharding technology to prevent data leakage or tampering.

The Corex Token network not only guarantees the authenticity of compute contributions, but also provides a trustworthy and traceable underlying infrastructure for global AI model training, laying a solid technical foundation for the development of the decentralized AI ecosystem.