2.1 Consensus Mechanism: Towards Efficient Proof of Computation

While PoS offers an improvement over PoW by reducing direct computational costs and energy consumption, it also introduces new challenges regarding resource allocation and incentive direction. This prompts us to rethink how to define a consensus and governance mechanism that is both efficient and fair in a decentralized network. Origins Network returns to the core question of a crypto project: how to effectively and reliably distinguish the majority of honest nodes from a minority of malicious ones. We propose a novel consensus algorithm that integrates key design philosophies from both PoW and PoS, aiming to create a hybrid mechanism that balances network security, resource efficiency, and fair incentives. This mechanism embeds a key principle of PoS: a block or chain is considered valid only when it obtains the endorsement (signature) of the majority of participants (represented by their weight).

Traditional native staking models (like Solo Staking) set multiple barriers for users in terms of capital threshold, reward stability, operational complexity, and learning costs. If one wishes to become a native staking node, it typically requires meeting specific hardware requirements, holding a minimum quantity of tokens, and accepting a lock-up period.

In the Proof of Work mechanism, miners compete for the right to create new blocks by stacking computing power (hashrate). In Origins Network's improved model, we reconfigure the dimension of competition. In addition to changes in the consensus mechanism, Origins Network also transitions block production time from a variable interval to a fixed cadence, introducing two time units: Slot and Epoch. A Slot is fixed at 12 seconds, and an Epoch is fixed at 6.4 minutes. An Epoch contains 32 Slots, meaning the network targets producing a block every 12 seconds and completing an Epoch cycle containing 32 blocks every 6.4 minutes.

Nodes obtain the qualification to become validators by staking $OR tokens. Subsequently, the Origins Network uses a Verifiable Random Function (VRF) algorithm to randomly select block proposers from all validators. The producer of each block is randomly chosen. Furthermore, within each Epoch, the Origins Network uniformly and randomly distributes all validators across different Slots, forming a "Committee" of at least 128 validators for each block, responsible for the validation and finality confirmation of that block.

Unlike traditional passive staking validation nodes, the validation process in the Origins Network public chain is designed as a structured, time-limited computational sprint. During this period, hardware providers (nodes) utilize their idle computational resources to solve a series of verifiable, complex computational tasks designed for AI workloads within a specified time window. Each completed sprint generates a cryptographic proof, attesting to the effective computational workload contributed by the node. This proof establishes the node's "baseline voting weight" for the next governance cycle. This enforces the principle of "one compute unit, one vote," ensuring a node's influence in network governance and consensus is proportional to its actual contributed computing power, not merely its held capital.

This method spiritually returns to the roots of Proof of Work—emphasizing the contribution of computing power—but optimizes its application with a key insight. It recognizes that the purpose of computational tasks used to ensure network security and fairness can be efficiently achieved within a limited time window, rather than requiring continuous, uninterrupted execution. During this limited sprint period, nodes focus their computing power to generate valid computational proofs. These proofs are then used as the basis for determining their voting weight in the subsequent cycle. This approach cleverly frees up substantial computational resources outside the sprint periods, allowing them to be freely and conflictlessly used for executing meaningful practical tasks (such as AI model training and inference), thereby significantly improving the resource utilization efficiency and long-term sustainability of the entire decentralized AI network.