3. Technical Architecture
Decentralized Computing Network Platform Design
3. Technical Architecture
3.1 Decentralized Computing Network Platform Design
The cAIToken decentralized computing platform is built on globally distributed device nodes. Its core goal is to split AI training tasks from traditional centralized data centers into on-chain verifiable task units, executed by participating nodes to earn cAI rewards from the unique mining pool. The platform adheres to the principle "single pool, zero permissions, output only from real contribution," centering around the release logic of 80,000,000 pre-minted tokens.
The underlying architecture adopts a modular "Task Slicer + Distributed Execution Layer" design, decoupling task segmentation, gradient computation, verification, and reward calculation to enable high concurrency, multi-node collaboration, and heterogeneous computing power execution. All task execution produces on-chain verifiable "computing proofs" as the sole basis for mining pool rewards.
cAI platform supports GPUs, CPUs, edge devices, and other node types without requiring specialized mining farms. Device detection, weight allocation, task distribution, and reward settlement are fully transparent, ensuring fair recording and incentive for all computing contributions. The platform uses an EVM-compatible architecture, enabling high scalability and portability of pool, task, and verification contracts.
3.2 Compute-PoW + AI-PoW
cAIToken's consensus model centers on the unique mining pool output mechanism --- 100% of token release comes from two verifiable actions: PoW computing power mining and node verification rewards from staking.
(1) Compute-PoW (Basic Computing Output)
Nodes perform basic computation via GPU/CPU and submit verifiable computing proofs (PoC) to receive basic pool output. Output depends only on:
Real device computing power
Online duration
Task execution quality
Computing reputation (historical performance)
Meaningless computation is not rewarded; all verifiable results must be reproducible, auditable, and tamper-proof.
(2) AI-PoW (Training Task Output)
Nodes execute training fragments, gradient calculations, inference tasks, etc., validated through the Proof-of-Training model. Greater training contribution earns higher rewards from the pool.
(3) Stake-Boost (Staked Task Enhancement)
Nodes staking cAI gain task acceptance and verification weight, allowing them to:
Earn additional verification rewards
Increase task allocation priority
Enhance long-term contribution value
Staking does not directly produce tokens but increases task and verification weight, fully conforming to the second release method emphasized. All rewards come 100% from the pool with no other issuance source.
3.3 Mining Pool Contract and Token Release Protocol
The mining pool contract is the ultimate core of cAIToken, adhering to three principles: unique, zero permissions, full-process transparency.
(1) Single Pool Architecture
80,000,000 cAI are minted once before mainnet launch and written entirely into the pool contract;
No team reserve
No private sale allocation
No investor share
No token control entry
This forms the basis of cAI's trustworthy distribution mechanism.
(2) No Admin Keys
The pool has renounced all administrative rights:
Cannot mint
Cannot pause issuance
Cannot modify release rules
Cannot adjust reward ratios manually
On-chain rules are permanent once written and can only be upgraded by DAO governance in future versions.
(3) Dual Release Logic (100% from Pool)
PoW: Output from real computing contributions
Staked Nodes: Earn ecological rewards as task executors/validators
Smart contracts calculate final rewards based on task execution, verification quality, computing reputation, and stake weight.
3.4 Security and Privacy Protection Mechanism
As the foundational infrastructure hosting global training tasks, cAIToken's security framework focuses on four dimensions: verification, anti-fraud, isolation, and transparency.
(1) Pool Security
Tokens have renounced permissions
Reward release is fully automated
No one can pause, modify, or intercept
Reward calculation and distribution are fully auditable
(2) Computing Power Authenticity Protection
Behavior analysis, replayable task verification, and GPU fingerprinting detect:
Fake computing submissions
Simulated GPUs
Replayed tasks
Forged gradients or inference results
Detected violations are automatically penalized, downgraded, or removed.
(3) Node Isolation and Reputation System
Each node has an on-chain reputation profile recording:
Successful tasks
Failure rate
Reward receipt
Stake status
Anomaly judgments
Higher reputation yields higher task priority.
(4) Privacy Protection
Training data is executed locally on nodes only, not uploaded or broadcast. Only the following is stored on-chain:
Parameter hashes
Contribution proofs
Verification results
User device information, data content, and model weights remain private.