L1, L2 or L1-L2 architecture?

Blockchains based solely on L1 have their own mechanisms of producing blocks, while blockchains comprising both L1 and place most of the utility/core infrastructures on L1 for efficiency and move most token logistics and value storage to the L2 layer. This setup often relies on many other well-known blockchain ecosystems to serve a wider range of customers and investors. L2 solutions typically embed their program logic and database into smart contracts within mainstream ecosystems. These projects integrate their logic entirely into smart contracts, complemented by a frontend framework connected to the backend contracts.

Architecture	Mainnet	Transaction fees on-chain	Transaction fees off-chain*	Supported tokens	Stablecoin integration?	Upgradability
L1	-	can be arbitrarily small	0	local	no	difficult
L2	Ethereum	0.0004 units	0	ETH&ERC20	yes	medium
	BSC	0.000075 units	0	BNB&BEP-20	yes	medium
	Tron	0.027 units	0	Tron&TRC20	yes	medium
L1-L2	Ethereum	0.0004 units	can be arbitrarily small	ETH&ERC20	yes	easy
	BSC	0.000075 units	can be arbitrarily small	BNB&BEP-20	yes	easy
	Tron	0.027 units	can be arbitrarily small	Tron&TRC20	yes	easy

The table summarized the performance matrix of different architecture designs. L1 ecosystems typically have their own databases and block production mechanisms. All transactions and associated state changes occur on-chain, using their local utility coins/tokens. Transaction fees ε can be set to very small values, as seen in the Tron network. However, once initiated, such blockchains can't easily be halted, and upgrades to core functions can be challenging. Such upgrades often necessitate a hard fork by miners or validators, which requires extensive communication between various parties to adopt a new protocol at a predetermined block height. In our effort to integrate web3 infrastructure with the AI marketplace, we need a setup that allows for ongoing system updates and feature additions without disrupting the network's assets or user experience. Building everything on L1 may not be the optimal solution.

L2 solutions, on the other hand, place all their core logic on a specific public mainnet, eliminating off-chain costs. All activities occur on-chain through contract calls to the mainnet. Assisted by oracles, L2 ecosystems span a wide range of areas including DeFi, Gaming, and NFT Marketplaces. Upgrades in L2 are typically handled using the upgradable contract paradigm, where contract updates are achieved by redirecting the proxy contract pointer. However, given that AI models and product upgrades cannot be fully migrated on-chain, this architecture is not suitable in the given context.

To effectively harness the potential of this decentralized network for web3 and AI marketplace merging, a two-layer L1-L2 architecture is introduced. The on-chain component (SC) records the value flow within the network, while the off-chain component (exec) comprises a set of protocols operating on the distributed network where utilities are executed. By seamlessly integrating the on-chain functionality with the diverse off-chain services provided, the system can achieve the robustness and upgradability that traditional Layer 1 solutions often lack. In the L1-L2 design, protocols and infrastructures mainly operate off-chain within the decentralized network, while token utilities like transfer and withdrawal function on Layer 2 of mainstream blockchains such as BSC or Polygon. This configuration enables the system to regularly update with new features and utilities, all while preserving the network's assets and the user experience. In the AI marketplace, the core module can be designed in L1 to ensure easy upgradability. Participants' databases can be distributed between L1 and L2 by placing their assets in L2 and conducting transactions in L1, thereby increasing efficiency and reducing costs. Protocols such as Chainlink and Proof of Training (POT) also adopt the L1-L2 architecture.