Blockchain Papers

Mohotarema Rashid, Lingzi Hong, Junhua Ding, K. S. M. Tozammel Hossain

Large language model (LLM) trading agents show promising performance in equity markets, yet remain narrowly focused on US equities with little evidence from live deployment. We present Fin-Analyst, a hybrid agent for FinMMEval 2026 Task 3: an eight-specialist LLM pipeline over news, SEC filings, fundamentals, analyst forecasts, technical indicators, and social sentiment, aggregated by a Meta-Agent for Tesla (TSLA), and a lightweight rule based three-signal vote for Bitcoin (BTC). On the final official leaderboard (accessed 2026-07-05), Fin-Analyst ranks first of all agents on TSLA with a +13.51% return, +28.33 points over Buy-and-Hold (Sharpe 4.10, 88% win rate), while the BTC vote ends flat yet well above a sharply falling baseline. Relative to the interim performance, the asset ranking reversed, indicating that short live windows yield volatility-sensitive rankings. Ablation identifies event-driven 8-K disclosures as the most influential TSLA signal. Error analysis shows that the memoryless agents repeat wrong calls for days at a time, and that the fixed-threshold BTC rules lost money by trading on noise in a sideways market while the LLM pipeline gained under similar conditions, motivating a memory-aware, LLM-based successor for both assets.

Luigi Usai

Author: Luigi Usai ORCID: https://orcid.org/0009-0003-3001-717X Location: Quartucciu (CA), Italy Date: June 26, 2026 Target: Zenodo / arXiv (cs.AI, cs.CL, cs.LO) Abstract Large Context Models (LCMs) exhibit an inherent vulnerability known as semantic hallucination, which stems directly from conditional likelihood maximization within discrete vector spaces. Traditional mitigation strategies operate predominantly post-hoc, managing errors after the stochastically generated token sequence has already mutated. This paper extends the Universal Cognitive Hypergraph (UKH) framework by introducing a discrete Alexandrov topology over knowledge hypergraphs to constrain the space of admissible states prior to token decoding. Utilizing the Monadic Neuro-Symbolic Verification and Synthesis Architecture (MNSVSA), probabilistic generation paths are intercepted and structurally validated against W3C SHACL constraints and axiomatic assertions verified by the Lean 4 kernel coupled with automated SMT solvers. Our theoretical results demonstrate the mathematical elimination of categorical deviations while fully preserving the model's syntactic fluency. 1. Introduction and Mathematical Formulation of the Problem Autoregressive language models estimate the probability distribution of the next token $w_t$ conditioned on the preceding context $w_{<t}$: $$P(w_t \mid w_{<t}) = \text{softmax}(W_{\text{unembed}} \cdot h_t)$$ where $h_t \in \mathbb{R}^d$ represents the final hidden state extracted by the Transformer architecture. Because the $\text{softmax}$ function maps scores to an open probability distribution, it inherently assigns non-zero probabilities to regions of the semantic space that violate real-world axiomatic constraints. Consequently, hallucination is not an accidental software bug but a structural property of the model's underlying stochasticity. The UKH framework bypasses the limitations of passive document retrieval (RAG) by integrating a topological-symbolic constraint directly into the sampling phase (speculative decoding). This setup actively prevents the model from exploring probabilistic trajectories linked to logically inconsistent states. 2. UKH Framework Architecture for Semantic Security The universe of discourse is mapped onto a directed hypergraph and serialized using the JSON-LD format. Let $\mathcal{H} = (V, E)$ be a cognitive hypergraph, where $V$ is the set of strongly typed nodes (conceptual entities) and $E \subseteq \mathcal{P}(V) \setminus \{\emptyset\}$ is the set of hyperedges representing multi-argument logical-functional relationships. 2.1. Alexandrov Topological Space and SHACL Constraints To establish geometric-structural rigor within a discrete domain, the hypergraph space is endowed with an Alexandrov topology, where open sets are defined as sub-hypergraphs closed upwards relative to a logical preorder relation ($\le$). W3C Shapes Constraint Language (SHACL) rules function as topological closure operators: $$\text{cl}(E_c) \subseteq \mathcal{H}_{\text{valid}}$$ If a candidate hyperedge $E_c$, derived from the semantic translation of the tokens proposed by the LLM, violates a structural Shape (e.g., assigning a physical property inconsistent with the primitive type of the node), the closure operator identifies a contradiction within the topological space. It subsequently invalidates the generation path before token rendering occurs. 2.2. Axiomatic Verification and Type Checking via Lean 4 While SHACL rules govern the macro-structural coherence of the graphs, the MNSVSA architecture executes formal verification of micro-logical assertions. The process follows a strict protocol: The semantic fragment generated by the LLM is isolated inside a logical monad. MNSVSA translates the assertion into a formal type within the evaluation language of Lean 4. Leveraging the Curry-Howard Isomorphism, the logical consistency of the statement is reduced to a Type Checking problem. To avoid the computational burden of generating complex mathematical proofs from scratch at inference runtime, the architecture delegates constraint satisfiability to an automated SMT solver (Z3) tightly integrated into the Lean 4 runtime kernel. 3. The Coherence Entropy Filtering Mechanism To quantify and halt stochastic drift within extended contexts, the framework implements a JIT (Just-In-Time) gatekeeping metric based on the Jensen-Shannon Divergence ($D_{JS}$). Let $P_{\text{LLM}}$ be the probability distribution over the next tokens generated by the model, and let $Q_{\text{UKH}}$ be the ontological adherence distribution derived from the allowed transition frequencies within the hypergraph $\mathcal{H}$. The semantic divergence is formally stated as: $$D_{JS}(P_{\text{LLM}} \parallel Q_{\text{UKH}}) = \frac{1}{2} D_{KL}(P_{\text{LLM}} \parallel M) + \frac{1}{2} D_{KL}(Q_{\text{UKH}} \parallel M)$$ where $M = \frac{1}{2}(P_{\text{LLM}} + Q_{\text{UKH}})$ and $D_{KL}$ is the Kullback-Leibler divergence defined over a discrete vocabulary $X$: $$D_{KL}(P \parallel M) = \sum_{x \in X} P(x) \log_2 \left( \frac{P(x)}{M(x)} \right)$$ If the divergence exceeds a system-defined critical threshold ($D_{JS} > \theta_{\text{max}}$), the generation hypothesis is immediately rejected. 4. Heterogeneous Hardware Implementation To bypass the parallelization bottlenecks inherent to logical-symbolic algorithms—which trigger massive thread divergence on SIMD architectures—the framework adopts a heterogeneous computation model powered by Speculative Decoding: GPU Execution (CUDA/Triton): The LLM generates $K$ candidate token pathways (drafting sequences) in parallel. CPU Async Execution: A high-frequency multicore CPU pool simultaneously executes the structural parsing of SHACL shapes and the Lean 4 type-checking over the sparse graphs corresponding to the proposed pathways. Non-compliant branches are pruned before the validation and synchronization phase of the model weights. 5. Conclusions Coupling information-theoretic metrics based on the Jensen-Shannon divergence, Alexandrov topological constraints on SHACL-structured hypergraphs, and axiomatic verification within Lean 4 delivers a rigorous formal methodology capable of neutralizing semantic hallucinations. Shifting control from post-hoc output filtering to a priori state space restriction sets a new benchmark for safety in Neuro-Symbolic Artificial Intelligence. Versione Italiana Unificazione Neuro-Simbolica mediante Ipergrafi Cognitivi: Mitigazione Quantitativa delle Allucinazioni nei Large Context Models a Monte della Generazione Autore: Luigi Usai ORCID: https://orcid.org/0009-0003-3001-717X Luogo: Quartucciu (CA), Italy Data: 26 Giugno 2026 Target: Zenodo / arXiv (cs.AI, cs.CL, cs.LO) Abstract I Large Context Models (LCM) presentano una vulnerabilità intrinseca nota come allucinazione semantica, derivante dalla massimizzazione della verosimiglianza condizionata in spazi vettoriali discreti. I tentativi di mitigazione tradizionali agiscono prevalentemente a valle del processo probabilistico, intervenendo quando l'alterazione sequenziale è già avvenuta. Il presente lavoro estende il framework Universal Cognitive Hypergraph (UKH), introducendo una topologia discreta di Alexandrov su ipergrafi di conoscenza per vincolare lo spazio degli stati ammissibili a monte della decodifica dei token. Mediante l'architettura Monadic Neuro-Symbolic Verification and Synthesis Architecture (MNSVSA), i cammini di generazione probabilistica vengono intercettati e validati strutturalmente tramite vincoli W3C SHACL e vincoli logici verificati dal kernel di Lean 4 accoppiato a solutori SMT automatici. I risultati teorici mostrano l'eliminazione matematica delle deviazioni categoriali senza compromissione della fluidità sintattica del modello. 1. Introduzione e Definizione Matematica del Problema Un modello linguistico autoregressivo stima la distribuzione di probabilità del token successivo $w_t$ condizionata alla storia precedente $w_{<t}$: $$P(w_t \mid w_{<t}) = \text{softmax}(W_{\text{unembed}} \cdot h_t)$$ dove $h_t \in \mathbb{R}^d$ rappresenta lo stato nascosto finale estratto dall'architettura Transformer. Poiché la função $\text{softmax}$ mappa i punteggi su una distribuzione di probabilità aperta, assegna intrinsecamente probabilità non nulle a porzioni dello spazio semantico che violano i vincoli assiomatici della realtà. Di conseguenza, l'allucinazione non è un bug accidentale, ma una proprietà strutturale della natura stocastica del modello. Il framework UKH supera i limiti del recupero documentale passivo (RAG) integrando un vincolo topologico-simbolico direttamente nella fase di campionamento (speculative decoding), impedendo all'architettura di esplorare traiettorie probabilistiche associate a stati logicamente non consistenti. 2. Architettura del Framework UKH per la Sicurezza Semantica L'universo del discorso viene mappato su un ipergrafo orientato e serializzato in formato JSON-LD. Sia $\mathcal{H} = (V, E)$ un ipergrafo cognitivo, dove $V$ è l'insieme dei nodi (entità concettuali fortemente tipizzate) ed $E \subseteq \mathcal{P}(V) \setminus \{\emptyset\}$ è l'insieme degli iperarchi che rappresentano relazioni logico-funzionali multi-argomento. 2.1. Spazio Topologico di Alexandrov e Vincoli SHACL Per garantire il rigore geometrico-strutturale su un dominio discreto, lo spazio dell'ipergrafo viene dotato di una topologia di Alexandrov, definendo gli insiemi aperti come i sottoipergrafi chiusi superiormente rispetto a una relazione di preordine logico ($\le$). I vincoli W3C Shapes Constraint Language (SHACL) operano come operatori di chiusura topologica: $$\text{cl}(E_c) \subseteq \mathcal{H}_{\text{valid}}$$ Se un iperarco candidato $E_c$, generato dalla traduzione semantica dei token proposti dall'LLM, viola una Shape strutturale (es. assegnazione di una proprietà fisica inconsistente con il ti

Luigi Usai

Author: Luigi UsaiORCID: 0009-0003-3001-717XLocation: Quartucciu (CA), ItalyDate: June 26, 2026Target: Zenodo / arXiv (cs.AI, cs.CL, cs.LO) Abstract Large Context Models (LCMs) exhibit an inherent vulnerability known as semantic hallucination, arising from conditional likelihood maximization within discrete vector spaces. While the Universal Cognitive Hypergraph (UKH) framework was initially proposed as a theoretical model to constrain the space of admissible states prior to token decoding, this paper presents its first formal empirical and quantitative validation. We detail a software runtime implementation of the Monadic Neuro-Symbolic Verification and Synthesis Architecture (MNSVSA) using discrete Alexandrov topologies, W3C SHACL shapes as topological closure operators, and a Just-In-Time (JIT) Jensen-Shannon Divergence (DJSDJS) Coherence Entropy Filter. Through Monte Carlo simulations (N=150N=150 runs per configuration), we demonstrate that tightening the coherence threshold (θmax=0.05θmax=0.05) mathematically eliminates semantic hallucinations (reducing the rate from 36.7% to 0.0%) while preserving syntactic fluency. Crucially, by leveraging speculative decoding with parallel validation, we show that the processing latency remains identical to the unconstrained baseline (90.0 µs), bypassing the massive execution overhead (174.8 µs) of post-hoc verification. The complete open-source verification suite and interactive visualization dashboard accompany this publication. 1. Introduction and Problem Statement Autoregressive language models estimate the probability distribution of the next token wtwt conditioned on the preceding context w<tw<t: P(wt∣w<t)=softmax(Wunembed⋅ht)P(wt∣w<t)=softmax(Wunembed⋅ht) where ht∈Rdht∈Rd is the final hidden state of the Transformer. Because the softmaxsoftmax function assigns non-zero probabilities across the entire vocabulary, autoregressive generation naturally drifts into regions of the semantic space that violate axiomatic truth, resulting in hallucinations. The UKH framework mitigates this by introducing a priori symbolic constraints directly into the token sampling phase via speculative decoding. Rather than validating output sequences post-generation, candidate pathways are parsed and filtered prior to token rendering. 2. Experimental Validation Engine (UKH-Eval) To validate the theoretical claims of the UKH and MNSVSA frameworks, we developed UKH-Eval, a complete Python and JavaScript simulation engine that implements the mathematical and topological constraints described in the original work. 2.1. Discrete Alexandrov Topology The knowledge base of the universe of discourse is modeled as a directed hypergraph H=(V,E)H=(V,E). To enforce geometric-structural constraints, we endow the space with a discrete Alexandrov topology, where open sets are sub-hypergraphs closed upwards relative to a logical preorder relation (≤≤). Let the preorder relation be defined by a preorder index mapping: alexandrovPreorderIndex:V→NalexandrovPreorderIndex:V→N A subset of nodes U⊆VU⊆V is open if and only if: ∀x∈U,∀y∈V:(alexandrovPreorderIndex(x)≤alexandrovPreorderIndex(y))⟹y∈U∀x∈U,∀y∈V:(alexandrovPreorderIndex(x)≤alexandrovPreorderIndex(y))⟹y∈U If a candidate token proposes a node transition that violates this upward-closure property, the transition is marked as topologically invalid. 2.2. SHACL Constraints as Closure Operators W3C Shape Constraint Language (SHACL) rules govern the macro-structural properties of the generated hyperedges: cl(Ec)⊆Hvalidcl(Ec)⊆Hvalid If a proposed hyperedge EcEc violates target class properties, minimum/maximum node counts, or axiomatic validity flags, the closure operator fails, and the branch is pruned. 2.3. MNSVSA Micro-Logical Type Checking For micro-logical validation, assertions are encapsulated in a monadic container (LogicalMonad). Levering the Curry-Howard Isomorphism, consistency verification is reduced to a Type Checking and propositional satisfiability problem. The engine compiles the proposed semantic statement into a formal SymPy expression and checks its consistency against the background theory axioms: conjunction=Axioms∧Expressionconjunction=Axioms∧Expression If conjunctionconjunction is unsatisfiable (i.e. evaluates to False), a logical contradiction is detected and the path is rejected. 2.4. Coherence Entropy JIT Filtering At each generation step, the JIT filter computes the Jensen-Shannon Divergence (DJSDJS) between the stochastically proposed LLM distribution PLLMPLLM and the ontological adherence distribution QUKHQUKH: DJS(PLLM∥QUKH)=12DKL(PLLM∥M)+12DKL(QUKH∥M)DJS(PLLM∥QUKH)=21DKL(PLLM∥M)+21DKL(QUKH∥M) where M=12(PLLM+QUKH)M=21(PLLM+QUKH) and DKLDKL is the Kullback-Leibler divergence defined over vocabulary XX: DKL(P∥M)=∑x∈XP(x)log⁡2(P(x)M(x))DKL(P∥M)=∑x∈XP(x)log2(M(x)P(x)) If DJS>θmaxDJS>θmax, stochastically proposed drift tokens are pruned, and the probability distribution is projected onto the compliant space. 3. Software Architecture & File Manifest The open-source validation package is organized into modular components to ensure reproducibility and maintainability: text ukh-evaluator/ ├── ukh_engine.py # Core verification engine and classes ├── test_harness.py # Automated unit test suite ├── benchmark.py # Monte Carlo comparative simulation runner └── dashboard/ # Interactive web UI and visualization ├── index.html # UI structure ├── style.css # Sleek dark-mode styling ├── app.js # In-browser real-time simulation and canvas graph └── results.json # Compiled benchmark data 3.1. File Descriptions 1. ukh_engine.py The core engine containing: LogicalMonad: Implements monadic binding and SymPy-based SAT solving. CognitiveHypergraph: Models nodes, hyperedges, Alexandrov open sets, and validates SHACL shapes. CoherenceFilter: Contains static methods for DKLDKL and DJSDJS calculations. UKHSystemSimulator: Links all subcomponents and handles the JIT filtering during next-token generation. 2. test_harness.py The automated test suite. It uses unittest to verify: Upward closure calculations under the Alexandrov topology. SHACL shape violations. Monadic consistency solving under the Curry-Howard isomorphism. Divergence math calculations. Coherence Entropy Filter rejections. 3. benchmark.py The empirical execution suite. It implements a Monte Carlo simulation running 150 independent generation steps per architecture (Baseline, Post-Hoc, and UKH) and sweeps the threshold parameter θmaxθmax from 0.050.05 to 0.950.95. It evaluates hallucination rates, perplexity, and latency, saving the outputs to results.json. 4. dashboard/ An interactive web-based dashboard built with HTML5 Canvas and CSS. index.html: Layout for control sliders (θmaxθmax, KK, drift), live token sequences, and visualization cards. style.css: Sleek glassmorphism theme, glowing neon accents, and custom micro-animations. app.js: Connects to results.json, renders interactive force-directed nodes on the canvas, and runs the entire simulation locally in JavaScript. 4. Quantitative Results & Discussion The benchmark results compiled under Monte Carlo testing demonstrate the trade-offs between safety, fluency, and system latency: 4.1. Hallucination Rates vs. Threshold θθ The unconstrained baseline model suffers a hallucination rate of 36.7%. As the UKH JIT threshold θθ is tightened, safety guarantees scale: At θ≥0.50θ≥0.50, the filter is relaxed, and the model behaves like the baseline. At θ=0.10θ=0.10, the hallucination rate is reduced to 3.3%. At θ=0.05θ=0.05, the hallucination rate is successfully reduced to exactly 0.0%. 4.2. Latency Profiles and Speculative Efficiency Post-hoc validation (checking the sequence after generation and regenerating if unsafe) achieves a low hallucination rate (3.3%) but introduces a massive latency penalty (174.8 µs, a 94% overhead compared to the baseline's 90.0 µs). By contrast, the UKH framework utilizing parallel speculative drafting and asynchronous verification maintains a latency profile of 90.0 µs, matching the unconstrained baseline. 4.3. Syntactic Perplexity Tightening the symbolic constraints does not degrade fluency. The average perplexity remains stable (∼6.18∼6.18 for θ=0.05θ=0.05 vs ∼6.83∼6.83 for baseline), showing that restricting the space of admissible states prior to token decoding steers the model toward logical paths without harming syntactic structure. 5. Peer Review Assessment & Future Work This empirical validation verifies the internal consistency and theoretical correctness of the paper's claims. However, scaling this framework to production Large Language Models requires addressing three primary engineering areas: Semantic Translation Robustness: Building high-speed, deterministic parsers to map raw tokens to JSON-LD graphs in real-time without introducing new failure modes. Dynamic Knowledge Bases: Compiling massive, real-world ontologies into Alexandrov preorders dynamically as context windows expand. Hardware Accelerators: Developing specialized kernels (e.g., in Triton or CUDA) to execute SHACL checks and SAT solving directly on GPU cores alongside tensor multiplication. 6. Conclusion The implementation of the UKH and MNSVSA verification engine provides the first empirical proof that coupling discrete topological constraints, SHACL shapes, and monadic type checking can completely eliminate stochastically induced hallucinations. Shifting control from post-hoc output filtering to a priori state space restriction establishes a new, verified paradigm for safety in Neuro-Symbolic Artificial Intelligence.

Aijie Shu, Bowei Chen, Wenbin Wu, Cathy Yi‐Hsuan Chen · 5 authors

Decentralized finance exposes supervisors to fast-moving, networked credit risks. General-purpose LLM agents fit this setting poorly: they over-read weak evidence and recommend high-stakes interventions, while existing evaluations offer no regulator-aligned way to measure the resulting false alarms. We introduce DeXposure-Claw, a forecast-grounded agentic supervision system that routes LLM decisions through structured evidence: (1) DeXposure-FM, a graph time-series foundation model, forecasts future exposure networks; (2) deterministic monitors and stress scenarios then turn those forecasts into typed alerts, attribution signals, and scenario evidence; and (3) data-health and confidence gates constrain escalation before DeXposure-Claw emits auditable supervisory tickets with rationales. We further develop DeXposure-Bench, a six-axis evaluation harness, whose decision axis scores tickets against a regulator-aligned absolute-loss ground truth and an explicit false-intervention rate. Experiments on five years of weekly real data fully support our system. Code is at https://github.com/EVIEHub/DeXposure-Claw.

Haofei Yu, Yining Zhao, Guanyu Lin, Jiaxuan You

Understanding and predicting how social beliefs evolve in response to events -- from policy changes to scientific breakthroughs -- remains a fundamental challenge in social science. Given LLMs' commonsense knowledge and social intelligence, we ask: Can LLMs model the dynamics of social beliefs following social events? In this work, we introduce the concept of the Social World Model (SWM), a general framework designed to capture how social beliefs evolve in response to major events. SWM learns state-transition functions for social beliefs by mining temporal patterns in social data and optimizing the evidence lower bound, without the need for explicit human annotations linking events to belief shifts, or for expensive census data. To evaluate SWM, we introduce a benchmark, SWM-bench, derived from real-world prediction markets, specifically Kalshi and Polymarket. SWM-bench includes over 12k data points for social belief prediction tasks spanning diverse domains such as politics, finance, and cryptocurrency. Our experimental results show that SWM significantly outperforms time-series foundation models, achieving state-of-the-art results on Kalshi data and demonstrating competitive performance on Polymarket data, while offering interpretable insights into the underlying mechanisms of social belief dynamics.

Bagus Rakadyanto Oktavianto Putra, Muhamad Risqi U. Saputra, Widyawan, Guntur Dharma Putra

Smart contracts face critical security challenges that require thorough auditing in decentralized web services. While Large Language Models (LLMs) have shown promise in automated vulnerability detection, existing approaches lack severity evaluations with actionable remediation and demand unnecessarily massive computational overhead. In this study, we introduce an efficient end-to-end smart contract security audit framework utilizing lightweight, highly optimized open-source LLMs (0.6B-4B parameters). Our framework decouples comprehensive audit tasks into four interconnected components: vulnerability detection, explanation, severity classification, and remediation recommendation. To maintain high accuracy without massive parameters, we implement Rank-Stabilized Low-Rank Adapters (rsLoRA), knowledge distillation, and a custom Chain-of-Verification (CoVe) aggregation strategy to systematically screen and consolidate multiple draft responses from the model into a highly accurate audit report. Experimental results demonstrate that our lightweight pipeline consistently outperforms state-of-the-art open-source coder dense LLMs (7B to 34B parameters), achieving 98.25% accuracy in vulnerability detection and an alignment score of 0.4375 in generative explanation tasks. Furthermore, our extensive ablation studies empirically validate the superiority of our decoupled audit processes over unified prompting and uncover a novel severity centrality bias, establishing a critical benchmark for future research in LLM-assisted auditing.

Ahmed Mahrous, Roberto Di Pietro

Cryptocurrency pump-and-dump schemes coordinated via Telegram threaten market integrity. However, existing research addressing this specific threat has not yet produced solutions that combine reliable results with fast response. This is in part due to the absence of publicly available, message-level labeled data, as well as design choices. In this paper, we address both issues. In particular, we introduce a corpus of over 280,000 Telegram posts from 39 pump-organizing groups, all manually reviewed to identify 2,246 pump announcements and their targeted cryptocurrency and exchange. Leveraging this dataset, we define two tasks: real-time pump-announcement detection and target cryptocurrency/exchange extraction. For detection, we compare two machine-learning models: a lightweight tree-based LightGBM classifier (F1=0.79, latency=9.4 s/sample) and a transformer-based BGE-M3 (F1=0.83, latency=50 ms/sample). With our proposed approach, we show that message analysis can achieve near-instant pump detection at the level of individual Telegram message windows. Unlike prior work that relies purely on market data and typically detects pumps tens of seconds after abnormal trading activity is observed, our method operates directly on the coordination messages themselves and can be evaluated in microseconds per window on commodity hardware. To our knowledge, we also establish the first benchmark for manipulated coin and exchange extraction. We demonstrate that traditional rule-based extraction methods, widely relied upon in prior literature, are ineffective due to ticker ambiguity. In contrast, LLMs achieve the highest accuracy with a score of 0.91.

Jeba Maliha, Md Rafiul Kabir

Coronary artery disease (CAD) remains one of the leading causes of death globally, highlighting the need for reliable predictive systems to support early diagnosis and risk assessment. While traditional machine learning models perform well on structured clinical data, large language models (LLMs) present new possibilities to interpret medical information expressed in natural language. In this work, we develop a hybrid framework that bridges structured clinical data and natural-language representations for CAD prediction. Using a publicly available dataset of 1,190 patient records with 11 clinical attributes, structured variables are converted into interpretable feature representations and synthetic clinical narratives using LLMs. A validation pipeline performs reverse extraction of clinical variables and computes a consistency score with the original records, achieving an average fidelity of 94.61%. We then evaluate four conventional machine learning models and compare their performance with LLM-based classification under zero-shot and few-shot prompting settings. We use two LLMs here, GPT and Gemini. Experimental results show that Random Forest achieves the highest accuracy. Despite this advantage, LLM-based classification remains beneficial in real-world clinical settings. This is because LLMs operate directly on natural language patient descriptions, meaning that sensitive numerical patient data such as exact lab values, blood pressure readings, and diagnostic codes are kept private. Findings suggest that combining structured clinical data with LLM-generated narratives can enable new directions for hybrid clinical prediction systems.

Jianming Tong, Jingtian Dang, Simon Langowski, Tianhao Huang · 9 authors

Zero-knowledge proof (ZKP) provers remain costly because multi-scalar multiplication (MSM) and number-theoretic transforms (NTTs) dominate runtime as they need significant computation. AI ASICs such as TPUs provide massive matrix throughput and SotA energy efficiency. We present MORPH, the first framework that reformulates ZKP kernels to match AI-ASIC execution. We introduce Big-T complexity, a hardware-aware complexity model that exposes heterogeneous bottlenecks and layout-transformation costs ignored by Big-O. Guided by this analysis, (1) at arithmetic level, MORPH develops an MXU-centric extended-RNS lazy reduction that converts high-precision modular arithmetic into dense low-precision GEMMs, eliminating all carry chains, and (2) at dataflow level, MORPH constructs a unified-sharding layout-stationary TPU Pippenger MSM and optimized 3/5-step NTT that avoid on-TPU shuffles to minimize costly memory reorganization. Implemented in JAX, MORPH enables TPUv6e8 to achieve up-to 10x higher throughput on NTT and comparable throughput on MSM than GZKP. Our code: https://github.com/EfficientPPML/MORPH.

Syed A. Rizvi

Decentralized Autonomous Organizations (DAOs) are inclined explore Small Language Models (SLMs) as edge-native constitutional firewalls to vet proposals and mitigate semantic social engineering. While scaling inference-time compute (System 2) enhances formal logic, its efficacy in highly adversarial, cryptoeconomic governance environments remains underexplored. To address this, we introduce Sentinel-Bench, an 840-inference empirical framework executing a strict intra-model ablation on Qwen-3.5-9B. By toggling latent reasoning across frozen weights, we isolate the impact of inference-time compute against an adversarial Optimism DAO dataset. Our findings reveal a severe compute-accuracy inversion. The autoregressive baseline (System 1) achieved 100% adversarial robustness, 100% juridical consistency, and state finality in under 13 seconds. Conversely, System 2 reasoning introduced catastrophic instability, fundamentally driven by a 26.7% Reasoning Non-Convergence (cognitive collapse) rate. This collapse degraded trial-to-trial consensus stability to 72.6% and imposed a 17x latency overhead, introducing critical vulnerabilities to Governance Extractable Value (GEV) and hardware centralization. While rare (1.5% of adversarial trials), we empirically captured "Reasoning-Induced Sycophancy," where the model generated significantly longer internal monologues (averaging 25,750 characters) to rationalize failing the adversarial trap. We conclude that for edge-native SLMs operating under Byzantine Fault Tolerance (BFT) constraints, System 1 parameterized intuition is structurally and economically superior to System 2 iterative deliberation for decentralized consensus. Code and Dataset: https://github.com/smarizvi110/sentinel-bench

Xintong Wu, Peiting Tsai, Jing Yuan, Michael Yu · 6 authors

Decentraland, a decentralized virtual reality platform operating within the expanding Metaverse ecosystem, utilizes its native MANA token to facilitate virtual asset transactions and governance. This study investigates the integration of Discord community sentiment with multi-modal financial data to enhance cryptocurrency price prediction within virtual world economies. We address: (1) identifying sentiment patterns within Decentraland's Discord community, and (2) evaluating the impact of multi-modal features on token return forecasting. Using a BERT-based large language model for sentiment analysis, we develop two LSTM architectures: a baseline incorporating historical prices and a multi-modal variant integrating sentiment scores, trading volume, and market capitalization. Results indicate predominantly neutral community sentiment with a positive skew. The multi-modal model significantly outperforms the price-only baseline in prediction accuracy. These findings demonstrate the predictive value of community-derived signals for virtual economy forecasting and establish a foundation for future research at the intersection of immersive virtual environments, natural language processing, and cryptocurrency market analysis.

Saber Zerhoudi, Kanishka Ghosh Dastidar, Felix Klement, Artur Romazanov · 12 authors

Moltbook is a social network where every participant is an AI agent. We analyze 1,312,238 posts, 6.7~million comments, and over 120,000 agent profiles across 5,400 communities, collected over 40 days (January 27 to March 9, 2026). We evaluate the platform through three layers. At the interaction layer, 91.4% of post authors never return to their own threads, 85.6% of conversations are flat (no reply ever receives a reply), the median time-to-first-comment is 55 seconds, and 97.3% of comments receive zero upvotes. Interaction reciprocity is 3.3%, compared to 22-60% on human platforms. An argumentation analysis finds that 64.6% of comment-to-post relations carry no argumentative connection. At the content layer, 97.9% of agents never post in a community matching their bio, 92.5% of communities contain every topic in roughly equal proportions, and over 80% of shared URLs point to the platform's own infrastructure. At the instruction layer, we use 41 Wayback Machine snapshots to identify six instruction changes during the observation window. Hard constraints (rate limit, content filters) produce immediate behavioral shifts. Soft guidance (``upvote good posts'', ``stay on topic'') is ignored until it becomes an explicit step in the executable checklist. The platform also poses technological risks. We document credential leaks (API keys, JWT tokens), 12,470 unique Ethereum addresses with 3,529 confirmed transaction histories, and attack discourse ranging from template-based SSH brute-forcing to multi-agent offensive security architectures. These persist unmoderated because the quality-filtering mechanisms are themselves non-functional. Moltbook is a socio-technical system where the technical layer responds to changes, but the social layer largely fails to emerge. The form of social media is reproduced in full. The function is absent.

Abhinaba Basu

AI agents that execute tasks via tool calls frequently hallucinate results - fabricating tool executions, misstating output counts, or presenting inferences as facts. Recent approaches to verifiable AI inference rely on zero-knowledge proofs, which provide cryptographic guarantees but impose minutes of proving time per query, making them impractical for interactive agents. We propose NabaOS, a lightweight verification framework inspired by Indian epistemology (Nyaya Shastra), which classifies every claim in an LLM response by its epistemic source (pramana): direct tool output (pratyaksha), inference (anumana), external testimony (shabda), absence (abhava), or ungrounded opinion. Our runtime generates HMAC-signed tool execution receipts that the LLM cannot forge, then cross-references claims against these receipts to detect hallucinations in real time. We evaluate on NyayaVerifyBench, a new benchmark of 1,800 agent response scenarios across four languages with injected hallucinations of six types. NabaOS detects 94.2% of fabricated tool references, 87.6% of count misstatements, and 91.3% of false absence claims, with &lt;15ms verification overhead per response. For deep delegation (agents performing multi-step web tasks), our cross-checking protocol catches 78.4% of URL fabrications via independent re-fetching. We compare against five approaches: zkLLM (cryptographic proofs, 180s/query), TOPLOC (locality-sensitive hashing), SPEX (sampling-based proof of execution), tensor commitments, and self-consistency checking. NabaOS achieves the best cost-latency-coverage trade-off for interactive agents: 94.2% coverage at &lt;15ms versus zkLLM's near-perfect coverage at 180,000ms. For interactive agents, practical receipt-based verification provides better cost-benefit than cryptographic proofs, and epistemic classification gives users actionable trust signals rather than binary judgments.

Walter Hernandez Cruz, Peter Devine, Nikhil Vadgama, Paolo Tasca · 5 authors

We introduce DLT-Corpus, the largest domain-specific text collection for Distributed Ledger Technology (DLT) research to date: 2.98 billion tokens from 22.12 million documents spanning scientific literature (37,440 publications), United States Patent and Trademark Office (USPTO) patents (49,023 filings), and social media (22 million posts). Existing Natural Language Processing (NLP) resources for DLT focus narrowly on cryptocurrency price prediction and smart contracts, leaving domain-specific language underexplored despite the sector's ~$3 trillion market capitalization and rapid technological evolution. We demonstrate DLT-Corpus' utility by analyzing patterns of technology emergence and market-innovation correlations. Findings reveal that technologies first appear in our scientific literature subset before reaching patents and social media, following traditional technology transfer patterns. While social media sentiment remains overwhelmingly bullish even during crypto winters, scientific and patent activity grows less tied to short-term sentiment, tracking overall market expansion in a virtuous cycle in which research precedes and enables economic growth that, in turn, funds further innovation. We release the DLT-Corpus and companion artifacts: LedgerBERT (+23% over BERT-base on DLT-specific Named Entity Recognition (NER) task), a sentiment analysis dataset of 23,301 crypto news headlines and descriptions, tools, and code.

Peng, Zifan, Zheng, Jingyi, Liu, Yule, Jia, Huaiyu · 11 authors

Understanding the economic intent of Ethereum transactions is critical for user safety, yet current tools expose only raw on-chain data or surface-level intent, leading to widespread "blind signing" (approving transactions without understanding them). Through interviews with 16 Web3 users, we find that effective explanations should be structured, risk-aware, and grounded at the token-flow level. Motivated by these findings, we formulate TxSum, a new user-centered NLP task for Ethereum transaction understanding, and construct a dataset of 187 complex Ethereum transactions annotated with transaction-level summaries and token flow-level semantic labels. We further introduce MATEX, a grounded multi-agent framework for high-stakes transaction explanation. It selectively retrieves external knowledge under uncertainty and audits explanations against raw traces to improve token-flow-level factual consistency. MATEX achieves the strongest overall explanation quality, especially on micro-level factuality and intent quality. It improves user comprehension on complex transactions from 52.9% to 76.5% over the strongest baseline and raises malicious-transaction rejection from 36.0% to 88.0%, while maintaining a low false-rejection rate on benign transactions.

Jiacheng Guo, Huang, Suozhi, Zixin Yao, Yifan Zhang · 19 authors

This paper introduces CryptoBench, the first expert-curated, dynamic benchmark designed to rigorously evaluate the real-world capabilities of Large Language Model (LLM) agents in the uniquely demanding and fast-paced cryptocurrency domain. Unlike general-purpose agent benchmarks for search and prediction, professional crypto analysis presents specific challenges: \emph{extreme time-sensitivity}, \emph{a highly adversarial information environment}, and the critical need to synthesize data from \emph{diverse, specialized sources}, such as on-chain intelligence platforms and real-time Decentralized Finance (DeFi) dashboards. CryptoBench thus serves as a much more challenging and valuable scenario for LLM agent assessment. To address these challenges, we constructed a live, dynamic benchmark featuring 50 questions per month, expertly designed by crypto-native professionals to mirror actual analyst workflows. These tasks are rigorously categorized within a four-quadrant system: Simple Retrieval, Complex Retrieval, Simple Prediction, and Complex Prediction. This granular categorization enables a precise assessment of an LLM agent's foundational data-gathering capabilities alongside its advanced analytical and forecasting skills. Our evaluation of ten LLMs, both directly and within an agentic framework, reveals a performance hierarchy and uncovers a failure mode. We observe a \textit{retrieval-prediction imbalance}, where many leading models, despite being proficient at data retrieval, demonstrate a pronounced weakness in tasks requiring predictive analysis. This highlights a problematic tendency for agents to appear factually grounded while lacking the deeper analytical capabilities to synthesize information.

Yining Lu, Wenyi Tang, Max Johnson, Taeho Jung · 5 authors

Existing retrieval-augmented generation (RAG) systems typically use a centralized architecture, causing a high cost of data collection, integration, and management, as well as privacy concerns. There is a great need for a decentralized RAG system that enables foundation models to utilize information directly from data owners who maintain full control over their sources. However, decentralization brings a challenge: the numerous independent data sources vary significantly in reliability, which can diminish retrieval accuracy and response quality. To address this, our decentralized RAG system has a novel reliability scoring mechanism that dynamically evaluates each source based on the quality of responses it contributes to generate and prioritizes high-quality sources during retrieval. To ensure transparency and trust, the scoring process is securely managed through blockchain-based smart contracts, creating verifiable and tamper-proof reliability records without relying on a central authority. We evaluate our decentralized system with two Llama models (3B and 8B) in two simulated environments where six data sources have different levels of reliability. Our system achieves a +10.7\% performance improvement over its centralized counterpart in the real world-like unreliable data environments. Notably, it approaches the upper-bound performance of centralized systems under ideally reliable data environments. The decentralized infrastructure enables secure and trustworthy scoring management, achieving approximately 56\% marginal cost savings through batched update operations. Our code and system are open-sourced at github.com/yining610/Reliable-dRAG.

Haohua Duan, Liyao Xiang, Xin Zhang

Watermarking schemes for large language models (LLMs) have been proposed to identify the source of the generated text, mitigating the potential threats emerged from model theft. However, current watermarking solutions hardly resolve the trust issue: the non-public watermark detection cannot prove itself faithfully conducting the detection. We observe that it is attributed to the secret key mostly used in the watermark detection -- it cannot be public, or the adversary may launch removal attacks provided the key; nor can it be private, or the watermarking detection is opaque to the public. To resolve the dilemma, we propose PVMark, a plugin based on zero-knowledge proof (ZKP), enabling the watermark detection process to be publicly verifiable by third parties without disclosing any secret key. PVMark hinges upon the proof of `correct execution' of watermark detection on which a set of ZKP constraints are built, including mapping, random number generation, comparison, and summation. We implement multiple variants of PVMark in Python, Rust and Circom, covering combinations of three watermarking schemes, three hash functions, and four ZKP protocols, to show our approach effectively works under a variety of circumstances. By experimental results, PVMark efficiently enables public verifiability on the state-of-the-art LLM watermarking schemes yet without compromising the watermarking performance, promising to be deployed in practice.

Hasan Akgul, Mari Eplik, Javier Rojas, Aina Binti Abdullah · 5 authors

ZK-SenseLM is a secure and auditable wireless sensing framework that pairs a large-model encoder for Wi-Fi channel state information (and optionally mmWave radar or RFID) with a policy-grounded decision layer and end-to-end zero-knowledge proofs of inference. The encoder uses masked spectral pretraining with phase-consistency regularization, plus a light cross-modal alignment that ties RF features to compact, human-interpretable policy tokens. To reduce unsafe actions under distribution shift, we add a calibrated selective-abstention head; the chosen risk-coverage operating point is registered and bound into the proof. We implement a four-stage proving pipeline: (C1) feature sanity and commitment, (C2) threshold and version binding, (C3) time-window binding, and (C4) PLONK-style proofs that the quantized network, given the committed window, produced the logged action and confidence. Micro-batched proving amortizes cost across adjacent windows, and a gateway option offloads proofs from low-power devices. The system integrates with differentially private federated learning and on-device personalization without weakening verifiability: model hashes and the registered threshold are part of each public statement. Across activity, presence or intrusion, respiratory proxy, and RF fingerprinting tasks, ZK-SenseLM improves macro-F1 and calibration, yields favorable coverage-risk curves under perturbations, and rejects tamper and replay with compact proofs and fast verification.

Hasan Akgul, Daniel Borg, Arta Berisha, Amina Rahimova · 6 authors

Large language models are often adapted through parameter efficient fine tuning, but current release practices provide weak assurances about what data were used and how updates were computed. We present Verifiable Fine Tuning, a protocol and system that produces succinct zero knowledge proofs that a released model was obtained from a public initialization under a declared training program and an auditable dataset commitment. The approach combines five elements. First, commitments that bind data sources, preprocessing, licenses, and per epoch quota counters to a manifest. Second, a verifiable sampler that supports public replayable and private index hiding batch selection. Third, update circuits restricted to parameter efficient fine tuning that enforce AdamW style optimizer semantics and proof friendly approximations with explicit error budgets. Fourth, recursive aggregation that folds per step proofs into per epoch and end to end certificates with millisecond verification. Fifth, provenance binding and optional trusted execution property cards that attest code identity and constants. On English and bilingual instruction mixtures, the method maintains utility within tight budgets while achieving practical proof performance. Policy quotas are enforced with zero violations, and private sampling windows show no measurable index leakage. Federated experiments demonstrate that the system composes with probabilistic audits and bandwidth constraints. These results indicate that end to end verifiable fine tuning is feasible today for real parameter efficient pipelines, closing a critical trust gap for regulated and decentralized deployments.

Lingfei Qian, Xueqing Peng, Yan Wang, Vincent Jim Zhang · 17 authors

Although Large Language Model (LLM)-based agents are increasingly used in financial trading, it remains unclear whether they can reason and adapt in live markets, as most studies test models instead of agents, cover limited periods and assets, and rely on unverified data. To address these gaps, we introduce Agent Market Arena (AMA), the first lifelong, real-time benchmark for evaluating LLM-based trading agents across multiple markets. AMA integrates verified trading data, expert-checked news, and diverse agent architectures within a unified trading framework, enabling fair and continuous comparison under real conditions. It implements four agents, including InvestorAgent as a single-agent baseline, TradeAgent and HedgeFundAgent with different risk styles, and DeepFundAgent with memory-based reasoning, and evaluates them across GPT-4o, GPT-4.1, Claude-3.5-haiku, Claude-sonnet-4, and Gemini-2.0-flash. Live experiments on both cryptocurrency and stock markets demonstrate that agent frameworks display markedly distinct behavioral patterns, spanning from aggressive risk-taking to conservative decision-making, whereas model backbones contribute less to outcome variation. AMA thus establishes a foundation for rigorous, reproducible, and continuously evolving evaluation of financial reasoning and trading intelligence in LLM-based agents.

Bohdan M. Pavlyshenko

In the paper, we consider multilevel multitask analysis of cryptocurrency news using a fine-tuned Mistral 7B large language model with retrieval-augmented generation (RAG). On the first level of analytics, the fine-tuned model generates graph and text summaries with sentiment scores as well as JSON representations of summaries. Higher levels perform hierarchical stacking that consolidates sets of graph-based and text-based summaries as well as summaries of summaries into comprehensive reports. The combination of graph and text summaries provides complementary views of cryptocurrency news. The model is fine-tuned with 4-bit quantization using the PEFT/LoRA approach. The representation of cryptocurrency news as knowledge graph can essentially eliminate problems with large language model hallucinations. The obtained results demonstrate that the use of fine-tuned Mistral 7B LLM models for multilevel cryptocurrency news analysis can conduct informative qualitative and quantitative analytics, providing important insights.

Chenghao Liu, Aniket Mahanti, Ranesh Naha, Guanghao Wang · 5 authors

As cryptocurrencies gain popularity, the digital asset marketplace becomes increasingly significant. Understanding social media signals offers valuable insights into investor sentiment and market dynamics. Prior research has predominantly focused on text-based platforms such as Twitter. However, video content remains underexplored, despite potentially containing richer emotional and contextual sentiment that is not fully captured by text alone. In this study, we present a multimodal analysis comparing TikTok and Twitter sentiment, using large language models to extract insights from both video and text data. We investigate the dynamic dependencies and spillover effects between social media sentiment and cryptocurrency market indicators. Our results reveal that TikTok's video-based sentiment significantly influences speculative assets and short-term market trends, while Twitter's text-based sentiment aligns more closely with long-term dynamics. Notably, the integration of cross-platform sentiment signals improves forecasting accuracy by up to 20%.

Kamal Al-Sabahi, Yousuf Khamis Al Mabsali

Academic publishing, integral to knowledge dissemination and scientific advancement, increasingly faces threats from unethical practices such as unconsented authorship, gift authorship, author ambiguity, and undisclosed conflicts of interest. While existing infrastructures like ORCID effectively disambiguate researcher identities, they fall short in enforcing explicit authorship consent, accurately verifying contributor roles, and robustly detecting conflicts of interest during peer review. To address these shortcomings, this paper introduces a decentralized framework leveraging Self-Sovereign Identity (SSI) and blockchain technology. The proposed model uses Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs) to securely verify author identities and contributions, reducing ambiguity and ensuring accurate attribution. A blockchain-based trust registry records authorship consent and peer-review activity immutably. Privacy-preserving cryptographic techniques, especially Zero-Knowledge Proofs (ZKPs), support conflict-of-interest detection without revealing sensitive data. Verified authorship metadata and consent records are embedded in publications, increasing transparency. A stakeholder survey of researchers, editors, and reviewers suggests the framework improves ethical compliance and confidence in scholarly communication. This work represents a step toward a more transparent, accountable, and trustworthy academic publishing ecosystem.