arXiv subjects

Joshua Tobkin

Publications and source records attributed to Joshua Tobkin.

6 recordsLinked to original sources

arXiv2026

Supra Cognitive Modes: A Routed Architecture for Agent Memory

Agent-memory workloads mix direct factual lookup, relation-chain and current-state reasoning, and broad synthesis over long histories. We describe Supra Cognitive Modes (SCM), an architecture that maps explicit or automatically selected per-query modes to retrieval and synthesis payloads over one shared ingest substrate. A frozen semantic classifier and runtime gates dispatch queries among fused lexical and dense lookup, graph or iterative multi-hop handling, and stratified long-form synthesis. The substrate combines multi-granularity embeddings, extracted triples, fact-version metadata, and optional asynchronous enrichments. We characterize the deployed configuration on three benchmarks: Long-term Conversational Memory (LoCoMo; n = 1,986), MemoryAgentBench (MAB; n = 3,671), and LongMemEval (n = 500). The reference run records 84.87% on LoCoMo factoid categories and 68.61% on adversarial abstention, 61.49% on MAB across two repetitions, and 86.00% on LongMemEval. A repository-backed reproduction produces similar aggregate scores and supports task- and mode-conditioned failure analysis. Raw baseline outputs, aligned end-to-end timing for LoCoMo and LongMemEval, and complete token ledgers are unavailable; stored rows also omit some final runtime decisions. The results characterize one implemented routed configuration and its diagnostic failure patterns, while source inspection verifies the per-query control interface and shared-substrate design. Causal routing effects, efficiency gains, and statistical significance remain outside the available evidence.

cs.AI↗

arXiv2025

Decentralised Multi-Manager Fund Framework

We introduce a decentralised, algorithmic framework for permissionless, multi-strategy capital allocation via tokenised, automated vaults. The system is designed to function analogously to a multi-strategy asset management company, but implemented entirely on-chain through a modular architecture comprising four interacting layers. The first, the capitalisation layer, composed of vaults that facilitate multi-asset deposits, tokenises investor participation, and specifies high level risk limits and admissible venues for deployment. The second, the strategy layer, enables the submission of strategies by human developers or autonomous agents, creating a decentralised marketplace governed by a validation mechanism incorporating adversarial and gamified elements. The third, the execution layer, operationalises strategy deployment using the host blockchain network's services. The fourth layer, the validated allocation layer, assesses and allocates capital among validated strategies, dynamically rebalancing toward those exhibiting superior risk-adjusted performance. In the framework, each admitted strategy acts as a manager for the "fund", encapsulated in a smart contract vault that issues transferable V-Tokens, conveying fractional ownership of the real-time portfolio operated by the vault. The system is designed to be open to participation by both human and AI agents, who collectively perform the roles of capital allocators, strategy developers, and validated allocators. The resulting structure is a self-regulating asset management ecosystem capable of decentralised, cooperative optimisation across traditional and digital financial domains. This framework is facilitated by a host chain network, which offers native automation and data oracle services enabling vault entities to autonomously operate on-chain, paving the way for being self sufficient in dynamic allocation of capital.

q-fin.GN↗

arXiv2025

Efficient Parallel Execution of Blockchain Transactions Leveraging Conflict Specifications

Parallel execution of smart contract transactions in large multicore architectures is critical for higher efficiency and improved throughput. The main bottleneck for maximizing the throughput of a node through parallel execution is transaction conflict resolution: when two transactions interact with the same data, like an account balance, their order matters. Imagine one transaction sends tokens from account A to account B, and another tries to send tokens from account B to account C. If the second transaction happens before the first one, the token balance in account B might be wrong, causing the entire system to break. Conflicts like these must be managed carefully, or you end up with an inconsistent, unusable blockchain state. Traditional software transactional memory (STM) has been identified as a possible abstraction for the concurrent execution of transactions within a block, with Block-STM pioneering its application for efficient blockchain transaction processing on multicore validator nodes. This paper presents a parallel execution methodology that leverages conflict specification information of the transactions for block transactional memory (BTM) algorithms. Our experimental analysis, conducted over synthetic transactional workloads and real-world blocks, demonstrates that BTMs leveraging conflict specifications outperform their plain counterparts on both EVM and MoveVM. Our proposed BTM implementations achieve up to 1.75x speedup over sequential execution and outperform the state-of-the-art Parallel-EVM (PEVM) execution by up to 1.33x across synthetic workloads.

cs.DC↗

arXiv2024

Moonshot: Optimizing Chain-Based Rotating Leader BFT via Optimistic Proposals

Existing chain-based rotating-leader BFT SMR protocols for the partially synchronous network model with constant commit latencies incur block periods of at least $2δ$ (where $δ$ is the message transmission latency). While a protocol with a block period of $δ$ exists under the synchronous model, its commit latency is linear in the size of the system. To close this gap, we present the first chain-based BFT SMR protocols with $δ$ delay between the proposals of consecutive honest leaders and commit latencies of $3δ$. We present three protocols for the partially synchronous model under different notions of optimistic responsiveness, two of which implement pipelining. All of our protocols achieve reorg resilience and two have short view lengths; properties that many existing chain-based BFT SMR protocols lack. We present an evaluation of our protocols in a wide-area network wherein they demonstrate significant increases in throughput and reductions in latency compared to the state-of-the-art, Jolteon. Our results also demonstrate that techniques commonly employed to reduce communication complexity$\unicode{x2014}$such as vote-pipelining and the use of designated vote-aggregators$\unicode{x2014}$actually reduce practical performance in many settings.

cs.DC↗

arXiv2024

Proof of Efficient Liquidity: A Staking Mechanism for Capital Efficient Liquidity

The Proof of Efficient Liquidity (PoEL) protocol, designed for specialised Proof of Stake (PoS) consensus-based blockchains that incorporate intrinsic DeFi applications, aims to support sustainable liquidity bootstrapping and network security. This concept seeks to efficiently utilise budgeted staking rewards to attract and sustain liquidity through a risk-structuring engine and incentive allocation strategy, both of which are designed to maximise capital efficiency. The proposed protocol serves the dual objective of: (i) capital creation by attracting risk capital efficiently and maximising its operational utility for intrinsic DeFi applications, thereby asserting sustainability; and (ii) enhancing the adopting blockchain network's economic security by augmenting their staking (PoS) mechanism with a harmonious layer seeking to attract a diversity of digital assets. Finally, the protocol's conceptual framework, as detailed in the appendix, is extended to encompass service fee credits. This extension capitalises on the network's auxiliary services to disperse incentives and attract liquidity, ensuring the network achieves and maintains the critical usage threshold essential for its sustained operational viability and progressive growth.

q-fin.GN↗

arXiv2023

DORA: Distributed Oracle Agreement with Simple Majority

Oracle networks feeding off-chain information to a blockchain are required to solve a distributed agreement problem since these networks receive information from multiple sources and at different times. We make a key observation that in most cases, the value obtained by oracle network nodes from multiple information sources are in close proximity. We define a notion of agreement distance and leverage the availability of a state machine replication (SMR) service to solve this distributed agreement problem with an honest simple majority of nodes instead of the conventional requirement of an honest super majority of nodes. Values from multiple nodes being in close proximity, therefore, forming a coherent cluster, is one of the keys to its efficiency. Our asynchronous protocol also embeds a fallback mechanism if the coherent cluster formation fails. Through simulations using real-world exchange data from seven prominent exchanges, we show that even for very small agreement distance values, the protocol would be able to form coherent clusters and therefore, can safely tolerate up to $1/2$ fraction of Byzantine nodes. We also show that, for a small statistical error, it is possible to choose the size of the oracle network to be significantly smaller than the entire system tolerating up to a $1/3$ fraction of Byzantine failures. This allows the oracle network to operate much more efficiently and horizontally scale much better.

cs.DC↗