PBS builder defection moves to the center of the discussion
ChainFeeds Research used its latest PRO issue to spotlight nuconstruct’s analysis of builder defection in proposer-builder separation, or PBS. In the current PBS setup, searchers hand bundles or trading strategies to builders, which gives builders visibility into the MEV embedded in those submissions. The report said the system does not force builders to execute according to the auction outcome, leaving room for them to copy a searcher’s strategy, reuse payloads, or capture the profit themselves.
nuconstruct framed the problem as a repeated game and asked whether reputation can keep builders honest. In theory, searchers can punish bad actors by moving future order flow elsewhere. In practice, the report argued, a single act of defection may be highly profitable, while proving that a builder copied or front-ran a strategy is often difficult. Under that logic, reputation works only when future revenue is large enough, cheating gains are relatively low, and misconduct can be detected easily.
Using libmev data, nuconstruct estimated that non-TEE builders could theoretically make about $26 million per year by reusing searcher information. If risks currently suppressed by BuilderNet’s TEE-based architecture are included, the total potential scale rises to about $49 million. ChainFeeds said this points to a structural issue in PBS rather than a narrow question of individual builder ethics, because builders can access valuable information before ordering is finalized.
The proposed fixes focus on mechanism design. The report listed several directions: using TEEs to limit a builder’s access to raw payloads, commitment-based settlement so builders must make verifiable commitments about ordering or execution outcomes first, and cryptographic ordering protections such as threshold encryption, commit-reveal, and verifiable delay functions, or VDFs, to keep transaction content hidden until ordering is set.
Bitcoin section covers NWC and fountain-code IBD
In the Bitcoin protocol section, ChainFeeds highlighted D-Central’s explanation of Nostr Wallet Connect, or NWC. The protocol lets applications interact with a user’s Lightning wallet without taking custody of funds and without direct access to private keys. The write-up framed the problem in practical terms: many apps want Lightning payments for tipping, subscriptions, paid content, in-game payments, or automated transactions, but asking every user to deposit funds into the app or import a wallet creates custody risk and poor user experience.
NWC separates the wallet from the app. Funds remain in the user’s chosen Lightning wallet, while the app communicates with that wallet through a Nostr relay to request invoices, send payments, check balances, or monitor payment status. During setup, the wallet generates an authorization string that the user passes to the app. The app can then operate only within the approved scope. Permissions can be limited by purpose, amount, and time. ChainFeeds described NWC as a connection-layer protocol rather than a new wallet or a new payment network.
The issue also referenced Bitcoin Optech Newsletter #413 and Lucas Lima’s work on applying fountain codes to initial block download, or IBD. The idea is to cut storage costs and let pruned nodes help other nodes reconstruct the chain without significantly increasing storage requirements. In the design described by ChainFeeds, the blockchain is split into fixed-length block groups by epoch, encoded into droplets with fountain codes, and sent together with block headers. A receiving node gathers enough droplets to decode the blocks and then uses the headers to verify validity.
Ethereum research: EIP-8287 and SPREAD
On Ethereum, ChainFeeds covered PLabs’ proposal for a privacy-native fungible token standard called pERC20 under EIP-8287. The goal is to define an ERC-20-like interface for the EVM while hiding balances, counterparties, and transfer amounts by default.
Instead of public account balances, the model represents tokens as encrypted UTXO-style notes. Each note stands for a private balance. Transfers use Groth16 zero-knowledge proofs and an Orchard-style model to prove spending rights, prevent double spends, and preserve value conservation without exposing the holder or the amount. At the same time, totalSupply remains publicly verifiable so hidden inflation cannot slip in unnoticed.
ChainFeeds noted that pERC20 is not compatible with ERC-20. It does not expose balanceOf, approve, allowance, or transferFrom, because those patterns reveal balances or relationships between users. The proposed IPERC20 interface instead includes methods such as transfer(PrivacyCall), mint, burn, totalSupply, and issuer. Transfers also do not emit the standard ERC-20 Transfer event. Wallets and indexers would watch events such as NoteAdded and NoteConfirmed to discover relevant notes. Each asset contract also maintains a cmxFrozenRoot, described as a Merkle root for a blacklist of frozen notes. When a user spends a note, the zero-knowledge circuit must prove that the note is not on the frozen list, allowing specific notes to be frozen without exposing identity or balances.
Another paper discussed in the issue came from Matheus Franco and co-authors, who proposed SPREAD as an optional extension compatible with GossipSub. The aim is to improve sender anonymity without a clear hit to propagation speed. ChainFeeds said standard GossipSub leaks timing and neighbor-relationship data. With only a small number of curious nodes, an attacker may observe the earliest delivery paths and link validators to peer IDs and IP addresses, opening the door to targeted DoS or economic attacks.
SPREAD combines two design elements. One is random walks inside local clusters to blur the true source of a message. The other is cluster formation based on virtual coordinates and geography or latency awareness, so cross-region propagation moves gradually through nearby clusters instead of relying on inefficient long-distance hops. Because the extension activates only when both sides support it, SPREAD can be deployed gradually while coexisting with current GossipSub nodes.
The experiments compared SPREAD with today’s GossipSub and with Dandelion++. Dandelion++ delivered the strongest anonymity, but the report said its latency is too high for low-latency settings such as Ethereum’s consensus layer. GossipSub remained reasonably fast but easier to deanonymize. SPREAD sat between the two, reducing the attacker’s success rate in identifying the sender while also showing better propagation efficiency than GossipSub and Dandelion++. The authors argued that if Ethereum slot time moves from 12 seconds to 6 seconds, the network layer will need both more speed and stronger resistance to deanonymization, making SPREAD a practical compromise.
MEV newsletter items and a DeFi lending paper
ChainFeeds also pulled several items from Flashbots’ The MEV Letter #146. One article, “ERC: Priority Update Registry (PUR)” by Quintus, vitaliy, and Tymur, proposed a draft contract standard for propAMM that uses a shared EVM registry so propAMM updates are easier for builders to order, verify, and analyze.
Another piece, “Lean Execution: a holistic approach to secure, efficient, adaptive, and resourceful execution throughput to scale the world-computer,” by Conall O’Reilly, examined Lean Ethereum’s execution-layer design and formalized Lean Execution as an execution-sharding architecture combining a virtual mempool, execution trees, and SNARK proofs for parallel sub-blocks.
The issue also cited nuconstruct’s “Builders’ Defection and Incentive Compatibility,” which argued that reputation alone cannot solve post-auction bidding and payload reuse in repeated PBS auctions, and pointed again to TEEs, commitment-based settlement, and encrypted order protection. It mentioned the podcast “Sergey Gorbunov on TEEs and the Arc Privacy Sector” from Zero Knowledge Podcast, as well as “All Core Devs - Testing (ACDT) #86, July 6, 2026,” hosted by Justin Traglia, which covered Glamsterdam devnet progress, specification changes, and client implementation, alongside an agenda post and an ACDT EL Breakout recording.
In the paper section, the newsletter summarized “Dynamic Interest Rate Discovery in Decentralized Finance: A Reverse Kelly Automated Market Maker for Risk-Adjusted Lending,” from authors affiliated with Woxsen University and Universitat de Girona. The paper examined the dependence of DeFi lending protocols on utilization curves and overcollateralization, then proposed a Reverse Kelly Automated Market Maker, or rkAMM, that applies the Kelly Criterion in reverse for credit-risk pricing.
According to ChainFeeds, rkAMM takes real-time probability-of-default inputs from an offchain explainable AI oracle, computes interest rates that can preserve target LP returns, and implements the design in Solidity smart contracts. Compared with the static utilization curves used by Aave and Compound, the authors said rkAMM can price risk at the individual-loan level more directly and improve capital efficiency for RWA and small and medium-sized enterprise financing.