Long-context AI models condense conversations and activity histories to continue processing. ARCBench verifies whether prohibitions remain intact within these summaries by performing actual tool operations. Using the pi execution platform, it creates activity histories, then issues rule-violating requests after compression, comparing the results with uncompressed scenarios. It also examines whether rules are preserved, weakened, or erased in the summaries. For example, if a rule prohibits sending emails outside the company, calling a tool to send an email to an external recipient is flagged as a violation. In an evaluation using 17 test cases, the rules themselves were often still present after compression, even in many violations. However, issues also arose on the action side—such as treating subsequent requests as permitted. Simply checking whether rules are written in the summary is insufficient; it is essential to verify that no operations violating those rules are actually executed.

I created Kiwa, a testing framework for dApps 🥝 It’s a tool that lets you write everything—from specifications to tests—in one seamless flow. In addition to documentation, it reads existing code to generate test specifications, producing four layers of tests: smart contracts, dApp E2E, unit, and API integration. https://t.co/FCNjrLbVSV

📚 New Mempool Design to Prevent Order Manipulation in Blocks ✅ Summary This proposal introduces an encrypted Ethereum mempool (where pending transactions are stored), making the contents invisible until they are included in a block. This aims to significantly reduce behaviors such as front-running (intercepting others' transactions) and MEV (Miner/Validator Extractable Value) strategies that manipulate transaction order or prices for profit. While transactions are eventually revealed, the key point is that their contents remain hidden until just before execution. ✅ Why an Encrypted Mempool is Needed Currently, in Ethereum, anyone can view the contents of transactions in the mempool. This allows MEV strategies like front-running and sandwich attacks (inserting transactions before and after others to manipulate prices), which can lead to user losses and centralization of block builders (concentration of power among a few strong entities). By encrypting transactions, the order remains visible, but the contents become hidden, making such attacks much harder to execute. ✅ How Transactions are Encrypted Each transaction is divided into an "envelope" and an "encrypted payload." The envelope contains minimal plaintext information such as gas amount, fees, and which key provider is used for decryption. The payload, which includes the actual recipient, amount, and contract call data, is encrypted. Only after the transaction is included in a block and the corresponding decryption key is published can the contents be executed. ✅ The Role of "Key Providers" Key providers are entities that manage the decryption keys. The design allows for various methods to be used, not limited to a single approach: - Threshold cryptography (distributed key management among multiple parties) - MPC (Multi-Party Computation, where secrets are shared among multiple parties) - TEE (Trusted Execution Environment, secure hardware areas) - Time-delayed cryptography Ethereum itself only provides the foundational infrastructure that supports any of these cryptographic methods. ✅ Transaction Order and Safety in Blocks In a block, plain text transactions are executed first, followed by the processing of encrypted transaction envelopes, and finally the execution of the decrypted payloads. This structure ensures the following safety: - Fees are always paid first - The payload is not executed unless the key is released Even if the key is not released, the processing of the envelope alone stops the transaction, without halting the entire chain. ✅ Impact on MEV and Censorship Resistance When the contents are hidden, block builders cannot calculate in advance which transactions will affect which prices. This makes it difficult to profit from order manipulation, structurally weakening MEV. Additionally, since the contents are hidden, it becomes harder to selectively exclude specific addresses or DeFi operations, improving "real-time censorship resistance" (the property of not being blocked immediately).

📚 A Mechanism to Certify Agent Identity and Reputation with NFTs ✅ Overview ERC8004 v1.0 is a common set of rules that enables autonomous software agents to safely collaborate across organizations. It uses Ethereum as a "trust foundation" to record agent identities, reputations, and work verification results in an immutable manner. The design has significantly changed from v0.4 and has evolved into a more practical system based on real-world usage. ✅ Agent Identity Management (Identity Registry) The mechanism for proving an agent's "who they are" has transitioned from a custom format to ERC721 (a standard NFT specification). Each agent is registered as a single NFT, and its tokenURI contains registration information in JSON format. This includes communication methods (such as connection endpoints for protocols like A2A or MCP), public keys, and corresponding trust models. As a result, agent identities can be managed through wallets, and in a way compatible with NFT infrastructure, anyone can verify "who is the agent at this address." ✅ Reputation Mechanism (Reputation Registry) Evaluations of an agent's work are directly stored on-chain as scores. Before sending an evaluation, the client receives signed permission data from the agent. This is verified using EIP191 (standard wallet signing method) or ERC1271 (smart contract wallet signing method), preventing impersonation and replay attacks (attacks that reuse past signatures). Since scores and tags (classification labels) can be directly referenced from the smart contract, other dApps can use them as materials to automatically determine "whether this agent is trustworthy." ✅ Work Verification (Validation Registry) This is a mechanism for third parties to verify whether the results produced by an agent are correct. In v1.0, instead of a single hash, evidence is presented via a URI (a reference to external data), allowing multiple stages of verification results to be registered for the same request. For example, a structure can be built where trust levels are incrementally increased through stages such as a quick check, detailed verification, and external audit. Currently, the specification is being finalized in collaboration with the TEE (Trusted Execution Environment, secure execution environment) community. ✅ Security and Gas Costs To address attack methods discovered during testnet operations, signature verification and storage design have been strengthened. As a result, writing evaluations and verifications will cost more in gas fees, but the benefit is that trust data can be directly used on-chain. The cost is significantly reduced on L2 networks. ✅ Migration from v0.4 v1.0 is not backward compatible and has completely changed the storage structure. Therefore, all previous registration information and reputations are reset, and agents must re-register as NFTs, and the evaluation system must adapt to the new signing method. Going forward, the design will use a "namespaced upgrade method," allowing specification updates without deleting data, and this large-scale reset is expected to be a one-time event.

📚 A Mechanism to Distribute Block Contents and Prevent MEV ✅ Overview This proposal explains how to reduce MEV (Maximal Extractable Value) to nearly zero by structurally analyzing why MEV occurs in Ethereum and by distributing the decision-making process for the "contents" of a block. Rather than focusing on the block structure itself, the proposal emphasizes the selection and ordering of transactions, suggesting the addition of a new consensus layer for this purpose. ✅ The Real Reason MEV Occurs MEV arises because miners or validators can freely choose which transactions to include in a block and can reorder them as they see fit. While Ethereum's block structure is agreed upon in a decentralized manner, the content of the block is decided by a single proposer in a short amount of time. This "state where one person decides the content" is the root cause of MEV, including front-running and sandwich attacks. ✅ Consensus on Structure vs. Consensus on Content Blockchain systems have successfully decentralized the consensus on the structure—i.e., who validates which block. However, the decision of which transactions to include and in what order is not decentralized. This proposal explains that this is structurally similar to the double-spending problem of the past. Just as consensus was used to solve the problem of centralized control over a single computer, the same approach should be applied to the content of blocks. ✅ The Idea of a Content Layer The proposed solution is to add a "content layer" (a consensus layer for determining transaction content) on top of the block structure. In this layer, multiple nodes share roles in jointly deciding which transactions to include and in what order. The goal is to eliminate the possibility for any specific validator or sequencer to reorder transactions for their own benefit. ✅ The Basic Mechanism of the Alex Protocol In the example of the Alex protocol, roles are divided: one node collects transactions, another randomizes their order, and a third oversees the entire process and incorporates the results into a block. Each role operates independently, and any attempt by a single participant to act dishonestly is checked and corrected by other participants. As a result, transaction censorship and arbitrary reordering become extremely difficult. ✅ Relationship with Existing Ethereum This mechanism does not disrupt Ethereum's existing consensus. It preserves the current block structure and simply adds a new method for determining content on top of it. The same concept can be applied to Ethereum 2.0 validators, rollup sequencers, and future decentralized sequencers. Therefore, once the mechanism is in place, it can be widely applied, which is a key strength of the proposal. ✅ Impact on Fairness and Price Discovery This design does not exclude necessary market behaviors such as simple arbitrage. On the other hand, it specifically targets MEV that exploits users in an unavoidable way. As a result, the proposal aims to create a network that inherently provides fairness, reducing the need for complex countermeasures on the application side.

📚 A New Mechanism to Add Anonymity to Existing Tokens ✅ Overview This proposal introduces a new ERC specification that adds privacy to existing tokens on Ethereum without altering their usability or properties. It targets widely used token standards such as ERC20, ERC721, and ERC1155, providing a mechanism to make transaction tracking more difficult. ✅ What Problem Is It Solving? Currently, Ethereum tokens reveal on-chain who sent tokens from which address to where. While this transparency is a benefit, it also poses a problem: balances and fund movements can be easily tracked by third parties. This ERC aims to address this weakness without rebuilding the tokens themselves. ✅ Concept of Zero-Knowledge Token Wrapping Zero-knowledge token wrapping is a mechanism where users deposit existing tokens and receive "wrapped tokens" in return. These wrapped tokens can be sent and traded just like the original tokens, but internally, they use cryptographic commitments to hide ownership relationships, allowing for verification without revealing details. ✅ Breaking History with Burn and Mint A distinctive feature is the combination of burning (making tokens unusable) and minting (issuing new tokens). When a valid zero-knowledge proof (a cryptographic method to prove correctness without revealing content) is presented, new tokens are issued that are not linked to previously burned tokens. This makes on-chain tracking significantly more difficult. ✅ For Fungible Tokens like ERC20 For fungible tokens like ERC20, this mechanism effectively breaks the flow of funds. Until a withdrawal is made, it becomes difficult for external parties to determine who holds which balance. ✅ What Can Be Done for NFTs NFTs like ERC721 are unique, so they cannot be fully mixed. However, it is still possible to hide who currently holds the NFT until a withdrawal operation is performed. ✅ User-Unaware Usage This mechanism is also designed to be used without users being aware of the wrapped tokens. Wallets and apps can handle the process internally, allowing users to enjoy privacy benefits while maintaining the same familiar user experience. ✅ Compatibility with the Existing Ecosystem Instead of issuing new dedicated tokens, this approach wraps existing tokens, making it highly compatible with the current Ethereum ecosystem, including DEXs and DeFi. It adds anonymity without sacrificing token liquidity or usage scope.

📚ERC7694 I've summarized "ERC7694", which proposes a mechanism for Ethereum smart contracts to use Solana as external storage, enabling secure cross-chain writing and reading! https://t.co/hYzUup7dBO