In briefGridSMR: Causal Compression for Faster Sharded Blockchains
SummaryThis paper presents GridSMR, a sharded blockchain that scales execution across shards and lets dependent cross-shard operations proceed within one block. Atomicity is limited to individual accounts, and cross-account work runs asynchronously, with consensus validating and committing the resulting schedule afterward. The authors report throughput of 1.07M requests/s with one validator and 193K committed requests/s with four validators. Latency for a 16-hop causal chain is reduced 8.2x compared with deferred execution.
In briefBlockchain Lifecycle Prediction: Predicting Dead Cryptocurrencies
SummaryThis paper examines whether cryptocurrency failure can be predicted from publicly available market data alone. An LSTM recurrent neural network was trained on 90-day sequences of daily price and estimated market capitalization for 82 assets (41 alive, 41 dead) from 2020 to 2026, using a strict chronological split. The best test-set ROC AUC reached 0.98, but dropped to between 0.59 and 0.65 on unseen data. The authors conclude that price and market-cap patterns carry useful signal, with failing assets showing gradual value erosion and rising volatility rather than sudden collapse.
In briefA Systematic Review of DeFi Vault Designs, Curator Controls, and Strategy Risks
SummaryThis paper systematizes decentralized finance (DeFi) vaults, smart-contract systems that pool deposits, run strategies, and issue share tokens. It presents a unified system model with formal definitions, then three taxonomies covering exposures and objectives, curator governance and accountability, and strategy execution patterns with their failure modes. The authors map representative production protocols onto these dimensions to support rigorous analysis and safer design of blockchain-based financial applications.
In briefReinforcement learning for flow-matching 3D mesh generation via dynamic homing
SummaryThis paper proposes Dynamic Homing Optimization (DHO), a reinforcement learning method for flow-matching 3D generation that redirects negative trajectories toward positive samples rather than only pushing them away. It uses Minimum-Cost Attractive Matching to assign each negative sample a positive target and Time-Aware Dynamic Correction to steer it with a remaining-time-aware velocity. Built on this, the Flow3D-Pro image-to-3D framework is reported to outperform DPO-, GRPO-, and NFT-style objectives and to produce higher-quality 3D geometry than existing mesh generation methods.
In briefSTAKE: A blockchain protocol that deters colluding majorities from double spending
SummaryThis paper proposes STAKE, a blockchain protocol that stays secure with only about one-third plus one honest participants, fewer than the usual two-thirds supermajority. Each participant must stake an amount large relative to their liquidity. The authors show that the probability of a coalition successfully executing a series of double-spending attacks drops exponentially with the number of attacks. They conclude that rational players do not collude with Byzantine ones even with a relatively low stake-to-liquidity ratio.
In briefBeyond Leaderboards: Cost, Latency and Token Efficiency of Small Model Ensembles
SummaryThis paper argues that evaluating LLMs in agentic workflows should go beyond accuracy to include reliability, cost, latency, and token efficiency. It uses an ensemble of small language models with a small-model judge feedback loop as a case study. On the 541-prompt IFEval benchmark, the best ensemble reaches 97.34% strict prompt accuracy, 5.81 points above the strongest standalone baseline, gpt-5.4, at lower cost.
Alexei N. Skurikhin, Emily M. Taylor, Nathan A. DeBardeleben
In briefPricing Bandwidth for Certified Transaction Dissemination in Blockchains
SummaryThis paper studies bandwidth fee mechanisms that price threshold-certified transaction dissemination before consensus, analogous to transaction fee mechanisms for computation. The authors model them as two-sided procurement auctions between users and validators, and prove limits on incentive compatibility, including that no nontrivial mechanism prevents profitable two-validator collusion. They give a greedy-route posted-price mechanism that is incentive compatible for individual users and validators, and propose an EIP-1559-style cross-block price-update rule with a locally stable fixed point under honest execution.
WhyIntroduces a new bandwidth fee framework with impossibility results, a feasible incentive-compatible mechanism, and stability analysis of cross-block pricing dynamics.
Eleftheria Fassman, Mahimna Kelkar, Benjamin Marsh et al.
In briefDAVINCI: A Decentralized Cryptographic Voting Protocol
SummaryThe paper presents DAVINCI, a decentralized end-to-end verifiable voting protocol that needs no trusted coordinator. It is built as a zero-knowledge rollup in which every election state transition is encoded as a circuit constraint, with Ethereum serving as a public bulletin board and data-availability layer. Ballot secrecy relies on threshold ElGamal with distributed key generation. The authors report that their open-source implementation supports large-scale elections and generates ballot proofs on commodity devices.
Pau Escrich, Marta Bellés-Muñoz, Roger Baig et al.
In briefFee mechanisms for pricing bandwidth to disseminate certified transactions
SummaryThis paper studies bandwidth fee mechanisms (BFMs) that price how transactions are shared among network nodes before consensus, when a threshold of nodes must certify them. It models this as a two-sided procurement auction between users and validators, and it is analogous to earlier transaction fee mechanisms but applies to communication instead of computation. The authors characterize limits on incentive compatibility and present a greedy-route posted-price mechanism that is incentive compatible for individual users and validators. They also propose an EIP-1559-style price-update rule and show its fluid approximation has a locally stable fixed point.
Eleftheria Fassman, Mahimna Kelkar, Benjamin Marsh et al.
In briefPINNing the pion: conformal deep learning for pion form factor and muon g-2 hadronic contribution
SummaryThis paper presents a physics-informed neural network built in a conformal z-plane to construct the pion electromagnetic form factor F_π(s) directly from first principles, rather than fitting phenomenological curves. Charge normalisation and Schwarz reflection are built into the architecture, while analyticity, dispersion relations, Watson's theorem and perturbative QCD asymptotics enter through the loss. The authors combine e+e− and τ-decay data and report estimates of the pion charge radius, ρ(770) pole parameters, and the two-pion contribution to the muon anomalous magnetic moment.
