Maryam Javaherian, Zafeiris Kokkinogenis, A. Pedro Aguiar
Sustaining cooperation in decentralized multi-agent reinforcement learning (MARL) remains challenging in social-dilemma environments, where agents operate under partial observability, unreliable communication, and incentives to free ride. This paper proposes a decentralized social mechanism for repeated local Public Goods Games that integrates structured commitment-based communication, bottom-up multi-dimensional reputation, and dual-timescale dyadic trust.Agents exchange messages about intended behavior, which are evaluated against realized actions to estimate commitment honesty. Reputation aggregates local evidence of cooperation, honesty, and behavioral consistency, while trust is updated at short and long timescales to balance responsiveness with robustness to noisy interactions. Without explicit social reward shaping, these signals influence learning through two coupled pathways: credibility-weighted message aggregation and trust-based participation filtering, which reshapes the effective local interaction pool used for payoff formation.We evaluate the framework across multiple public-good multipliers using a common topology-local payoff setting. Comparisons include a baseline, a communication-only configuration, an Only-Reputation configuration with hard reputation-based exclusion, participation-filtering variants, and single-timescale trust ablations. The Full Model improves the cooperation--payoff trade-off and maintains cooperation more consistently than reduced-mechanism configurations. Analyses of learning dynamics, across-seed variability, participation mechanisms, trust timescales, and threshold sensitivity indicate that improved cooperation is associated with combining credibility-aware communication, trust and reputation, and payoff-relevant participation filtering. These results suggest that socially grounded information and interaction mechanisms can support more stable cooperation in decentralized MARL without explicit reward shaping.
Decentralized prediction markets increasingly support trading in claims on elections, macroeconomic events, sports outcomes, and regulatory decisions, yet little is known about the microstructure through which information enters their prices. This study examines informed trading on Polymarket by reconstructing trade direction from raw on-chain OrderFilled settlement events and converting YES and NO token trades into a common YES-equivalent buy/sell measure. Using 189,623 market-day observations across 5,518 resolved binary contracts, we estimate market-level Probability of Informed Trading (PIN). We document economically meaningful informed-order-flow intensity: mean PIN is 0.193, median PIN is 0.186, and volume-weighted PIN is 0.192. PIN remains stable across restricted samples and is positively associated with model-independent order imbalance. Cross-sectional evidence shows that informed trading is shaped more consistently by market design and contract lifecycle than by raw volume: negative-risk markets and longer-lived contracts exhibit higher PIN, whereas volume effects are not robust.
Private transfers on public smart contract blockchains hide transaction values and private transfer links, but accountable privacy requires controlled auditability. Existing auditable zero‐knowledge transfer systems make per‐transaction audit information available to an authorized auditor, but their audit model may allow the auditor to inspect transactions beyond the flow connected to the authorized audit target. In such a model, granting audit capability for one investigation can expose unrelated transactions or allow tracing to continue farther than intended. In this paper, we propose token‐guided flow tracing for auditable zero‐knowledge smart contract transfers, enabling audit authorization to be scoped to a transaction flow and epoch range. The construction separates audit authorization from audit capability by introducing a tracing‐token manager and an auditor. The auditor can open audit information only when it holds the auditor secret key, an epoch secret key for an authorized epoch, and a tracing token associated with the target transaction flow. Direction‐specific tracing tokens confine tracing to the authorized direction, while epoch public keys and a binary key‐derivation tree support compact authorization of audit intervals. We analyze security through ledger indistinguishability, transaction nonmalleability, balance, audit correctness, and restricted audit authorization. Restricted audit authorization is established in an honest authorization model that assumes the tracing‐token manager and the auditor do not collude beyond explicit audit authorizations; under these assumptions, the auditor cannot open or trace transactions outside the authorized flow, direction, and epoch scope. Our implementation adds 16,349 transfer‐circuit constraints, about 178,000 gas to each accepted transfer, and 448 bytes to the serialized transfer transaction, showing that scoped auditability can be added with moderate on‐chain overhead.
Decentralized digital assets are characterized by pronounced volatility, non-linear dynamics, and complex temporal dependencies. These properties often render traditional econometric models and basic neural networks insufficient for long-horizon forecasting. Although volatility estimation techniques have matured, forecasting multi-horizon price movements remains a significant challenge due to non-stationarity and high-dimensional noise in these markets. To overcome these limitations, we introduce a hybrid architecture combining a One-Dimensional Convolutional Neural Network (1D-CNN) with an Independently Recurrent Neural Network (IndRNN). Designed for multi-horizon forecasting at 7, 30, and 90-day intervals, this framework processes high-dimensional cryptocurrency datasets, including market trends, technical indicators, transaction metrics, and on-chain variables. We systematically evaluate filter, wrapper, and embedded feature selection techniques (Forward, Backward, and Exhaustive) to eliminate redundancy and optimize computational cost. The 1D-CNN extracts spatial features, while the IndRNN captures long-term temporal dependencies without gradient degradation. Benchmark evaluations on Bitcoin (BTC), Ethereum (ETH), Dogecoin (DOGE), Bitcoin Cash (BCH), and IoT Chain (ITC) demonstrate the model’s effectiveness. With strict linear split validation simulating realistic future conditions, the model reduces Mean Absolute Error (MAE) by about 18\% and Mean Absolute Percentage Error (MAPE) by nearly 20\% compared to baselines such as LSTM, ANN, and SVM. In classification, the framework achieves up to 81\% accuracy on 90-day forecast horizons. Unlike resource-intensive Transformer models—which often struggle with limited data and abrupt market shifts—our lightweight architecture offers superior computational efficiency and practicality for trading, asset management, and quantitative risk modeling.
Empirical option backtests observe strategy returns only when all required contracts can be mapped and qualifying transactions appear within the measurement window. This creates endogenous strategy-sample formation before performance is measured. We reconstruct that process for a predefined registry of daily inverse Bitcoin-option strategies on Deribit. Across both study layers, 69,020 nominal template-date mappings yield 49,558 primary actual strategy-date positions after geometry checks and deduplication, and 18,457 of those positions complete within 60 minutes. In the primary Layer A sample, the unweighted mean of date-level completion rates is 40.9% (stationary date-block 95% interval, 37.0%–45.0%). A construct-validity diagnostic based on 355 provider-ID event-strategy matches, corresponding to 352 exact-signature clusters, shows that the public-tape index captures a related activity margin but is neither atomic package execution nor an investor fill. Completion varies across strategy families and market states. Moreover, completed-observation, full-calendar, inverse-probability-weighted, and augmented conventions can change backtest rankings and signs. Conventional inference does not identify a rule superior to the at-the-money benchmark, and the evidence supports no deployable trading claim. The contribution is a financial-inference framework that makes strategy formation, missing outcomes, dependence, and temporal transport explicit.
Bitcoin is the asset. Debt and preferred securities are the senior claims. MSTR common is the residual. Understanding that hierarchy is the key to understanding both the upside and the risk.
This paper argues that the next stage of AI architecture should not be conceptualized primarily as larger language models, but as systems optimized for contextual comprehension, explicit knowledge integration, inconsistency handling, consequence anticipation, and auditable explanation. Drawing on foundation-model research, critiques of meaning-through-form alone, causal inference, neuro-symbolic AI, retrieval-augmented generation, and agentic systems, it proposes an operational definition of understanding as the capacity to maintain a task-relevant semantic state, ground claims in evidence, reason over constraints, evaluate action consequences, and produce explanations that remain traceably coupled to decision processes. On this basis, the paper introduces understanding-centric, consequence-aware architecture patterns, together with evaluation metrics and experimental protocols spanning language, reasoning, retrieval, contradiction handling, and action tasks. It also situates the broader LUM research stream as a historical research-program label inside the text while deliberately removing scale from the title and from the core definition of understanding. The resulting agenda is intended as a rigorous bridge between LLMs, SLMs, neuro-symbolic systems, and enterprise decision intelligence.
We study the establishment of U.S. National Laboratories in the 1940s–1950s to estimate local spillovers from public research infrastructure. This setting allows us to causally identify such spillovers, for two reasons: 1) Lab sites were chosen largely for security and political reasons, rather than existing or potential innovative capability and 2) We identify runner-up locations using archival sources. We find several types of knowledge spillovers: Compared to control counties, Lab counties experience large and persistent increases in patenting by non-lab inventors; non-lab patents in the same county shift toward laboratories’ research fields and cite laboratory patents more frequently. Using newly digitized county data from 1936–1970, we find sustained increases in retail sales and household income. Linked 1940–1950 Census records show wage gains for pre-existing residents who remain in lab counties, with larger effects for college-educated workers. We find that cohorts exposed to laboratory establishment during school-age years attained more education, consistent with a human-capital channel. Spillovers arise despite extensive secrecy around early nuclear research, suggesting that co-location with public R&D can generate sizable local benefits even under restricted information flows. Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org .
Scientific analysis of the factual medical evidence in public view proves beyond reasonable doubt that 3 year-old Hannah’s untimely death in 2018 had been caused by an invisible head injury sustained in a playtime misadventure a full 15 hours before she suddenly collapsed. But a trained radiologist misrepresented the meaning of Hannah’s CT scan to fit a prejudiced opinion of child abuse causing immediate unresponsiveness, triggering similar errors by a pediatrician, then by police, then by Hannah’s innocent babysitter Lindsay Partin, then by an opthalmologist, then by a pathologist, and eventually by a judge and jury, an appeal court, and finally by the Supreme Court of Ohio. The sociological history of pediatrics sheds some light on how a qualified doctor could make such a dreadful error of judgment, and the science of the subconscious partly explains why Partin made a false confession of assault.
In the context of Russia's brutal invasion of Ukraine, an opportunity, previously seen as unrealistic, arose: The Republic of Moldova received an EU candidate status in June 2022. As the Chisinau administration is currently the most pro-European in the country's history, and considering the great mobilisation towards integration on Brussels's side, it is a very relevant endeavour to analyse Moldova's main opportunities and barriers. This paper looks at the European integration process within the Republic of Moldova, by analysing their historical evolution (both within the Soviet Union and post-Soviet).
This essay examines the growing crisis of proof in the age of artificial intelligence. As digital systems become more powerful, more automated, and more opaque, the capacity to establish basic facts such as origin, integrity, chronology, attribution, provenance, and institutional usability is becoming increasingly fragile. The paper argues that law cannot remain effective without a minimal evidentiary infrastructure. It proposes a structured framework for thinking about proof not merely as a procedural matter, but as a technical, legal, institutional, and political architecture. It develops the idea that generative AI does not create the evidentiary crisis, but accelerates and exposes it. In this context, transparency alone is insufficient if it remains purely declarative and cannot be independently verified. The essay therefore explores the need for portable, auditable, and proportionate proof layers capable of supporting rights, accountability, negotiation, and public trust without collapsing into generalised surveillance. AURA is introduced as a first minimal building block within this broader architecture: not as a total solution, but as a lightweight proof-of-origin and integrity layer designed to create verifiable fixed points in environments dominated by informational asymmetry. The essay ultimately argues that the next critical layer of the digital order may not be intelligence alone, but the ability to make certain facts provable again.
This study investigates how the number of large-stake lotteries in a questionnaire affects parameter estimation in Prospect Theory (PT) and Cumulative Prospect Theory (CPT). Using binary-choice data from Japanese university students, parameters are estimated via maximum likelihood estimation combined with a genetic algorithm. Twelve models are constructed by combining alternative value functions and probability weighting functions. The results show that exponential value functions yield stable parameter estimates across different questionnaire designs, whereas power value functions are sensitive to the number of large-stake lotteries. Fewer large-stake lotteries lead to higher estimated risk aversion in gains and higher risk seeking in losses. These findings highlight the importance of functional form selection and questionnaire design in empirical applications of prospect theory.
Solving continuous-time macroeconomic finance models is very challenging. Especially when modeling the dynamics of multiple assets, state variables can easily lead to the curse of dimensionality once their number increases slightly. On the other hand, without appropriate boundary conditions, traditional methods for solving such models face significant challenges in convergence (for example, using the iterative method provided by Brunnermeier and Sannikov (2016)). This paper proposes a two Lucas trees model, consisting of a crypto tree. A key feature of this crypto tree is its dividends are not exogenous but tied to the value of the crypto tree itself. It's an abstraction of Bitcoin in reality: Bitcoin's total supply is tightly capped, with new coins periodically mined. The value of newly mined coins is ultimately determined by the market value of Bitcoin. Another tree produces dividends with follows OU process As a probabilistic method for solving PDEs, Deep BSDE has two key advantages. First, it bypasses the complexity of finite difference methods. Second, it is not highly sensitive to boundary conditions, allowing one to solve the PDE within a region of the state space that has a high probability(mass), which is helpful to solve this crypto tree model.
We study whether macropolitical risk, as revealed by information leaders in prediction markets, drives the jump dynamics of Bitcoin and Ether. We identify leading oracle contracts using directed Transfer Entropy and construct a signed INFL index from their probability revisions. We then embed the INFL index as an exogenous excitation term in Heston-SVCJ-Hawkes models and use Bayesian MCMC to estimate the coupling coefficients. We find that the INFL index accounts for most signed jump-intensity variation, while positive cross-sign couplings are consistent with a de-risking pattern in which macropolitical shocks simultaneously elevate upside and downside jump intensities.
General Ledger (GL) reconciliation is a fundamental control in financial reporting, which is critical for accuracy, transparency and reliability of accounting records both in public and private sectors. Yet, the continued reliance on manual processes, outdated technology and inconsistent reconciliation standards have resulted in systemic weaknesses such as financial misstatements, regulatory exposure, operational inefficiency and trust among stakeholders. These deficiencies are exacerbated by increasingly intricate and internationally linked financial systems, which further increase the risks to both an institution's performance and overall stability. This study reinforces the idea that modernization is not an engineering luxury, but rather a strategic necessity. The research is lighting on automation, RPA, AI, blockchain and Cloud-based ERP platforms which enable a transformation of reconciliation from a periodic error prone task to a continuous proactive and highly reliable financial control. These results also underscore the importance of governance reform, data standardization and effective change management in sustaining gains from technological innovation. Barriers that are specific to sectors, for example lack of resources and procurement difficulties in the public sector or integration and cyber-security risks in the private sector require both a bespoke response with regulatory cooperation and cross-sector measures. Finally, up-to-date GL reconciliation systems are poised to strengthen organizational agility, increase transparency and secure the integrity of a financial ecosystem. By combining sophisticated technology and strong governance, institutions will be able to minimize systemic weaknesses, enhance transparency and construct a more sustainable and trusted financial reporting ecosystem.
This study investigates the effect of general ledger mastery on students’ ability to prepareatrial balance in accounting education. The general ledger plays a central role intheaccounting cycle, as it summarizes journalized transactions and directly determines theaccuracy of trial balance preparation. However, many students still experience difficultiesinlinking ledger procedures to trial balance construction, resulting in classification errors andimbalance. This research employed a quantitative approach with a correlational designinvolving undergraduate accounting students. Data were collected through performance-based tests, questionnaires, observations, and documentation, and analyzed using descriptivestatistics, correlation, and regression techniques. The results reveal a significant positiverelationship between general ledger mastery and trial balance preparation ability, indicatingthat stronger ledger competencies contribute to higher reporting accuracy. These findingsconfirm the sequential logic of the accounting cycle and highlight the importanceofstrengthening ledger instruction. The study concludes that systematic training inledgerprocedures is essential to enhance students’ financial reporting competence and recommendsintegrating scaffolded and technology-supported learning in accounting instructio.
The objective of this research is to investigate Triple Entry Accounting (TEA), an evolution from the traditional double entry bookkeeping. To start, the traditional double entry system provides internal consistency but is subject to information asymmetry, reconciliation challenges, and human fraud. This research aims to investigate whether a "single source of truth" for inter-corporate transactions can be achieved by utilizing a third entry in a blockchain system. The methodology used is a simulation created in Microsoft Excel. The research looks into the various advantages of the application of Triple-Entry Accounting. The research results show that a Triple-Entry system provides a 100% detection capability for quantitative discrepancies, reducing the "time to detection" from weeks to seconds. Additionally, the research demonstrates how AI can be used as a complementary tool to assist with continuous assurance. The research concludes that the move to a Triple-Entry system is not merely a technical evolution, but a shift to a system of "financial transparency" and "proof of solvency." With the elimination of the need for retrospective reconciliation, Triple-Entry Accounting opens the door to a dynamic reporting environment that will increase the integrity of global financial markets.
Hua Yang, Yongli Wang, Yantao Pan, Bin Liu · 6 authors
Diffusion Transformers (DiTs) are powerful backbones for visual pattern synthesis, but iterative denoising repeatedly applies full-token self-attention, limiting high-throughput recognition workflows. We propose drift-controlled Denoising-Aware Dynamic Token Pruning (DADP), a plug-and-play training-free inference framework for efficient DiT sampling. DADP scores tokens using normalized activation energy and centered-token deviation, allocates retention through a layer-timestep scheduler, and stabilizes token support with exponential moving average continuity. Selected tokens are processed as a compact sequence, after which their residual updates are scattered to the original grid while unselected tokens follow an identity bypass. The centered deviation is interpreted as an internal covariance-drift statistic, linking token selection to hidden-distribution preservation without modifying pretrained weights, conditioning paths, or output shapes. On a 28-layer DiT with 50 steps at $256\times256$, DADP has a mean retention ratio of 0.595, reduces attention FLOPs by 58.9\%, and yields $2.43\times$ attention-only and $1.33\times$ measured end-to-end speedups, with FID$_{\mathrm{DiT}}$ 4.5081. Under the conservative $512\times512$ budget, it retains 0.707 of tokens, reduces attention FLOPs by 43.6\%, and yields $1.77\times$ attention-only and $1.27\times$ end-to-end speedups, with FID$_{\mathrm{DiT}}$ 5.7828. At both resolutions, DADP improves dense-reference FID over fixed random and similarity pruning.
We present a Kaplan-Meier and Cox proportional-hazards survival analysis of 832,941 Solana pump.fun token launches with 24-hour graduation outcomes, observed continuously between 2026-05-08 and 2026-06-10. The pooled graduation rate is 0.198% (Wilson 95% CI [0.189%, 0.208%]), a 3.18x decline from the 0.63% rate reported by Marino et al. (2026) for September-October 2025. After excluding a four-day graduation-tracker warm-up at the start of our window and the partial last day, the steady-state rate is 0.207% (Wilson 95% CI [0.198%, 0.218%]); the structural effects we report below replicate within decimal-place noise under this exclusion. We formalise the windowed comparison as the Graduation Regime Windows (GRW) framework, name our 34-day window the RED-PUMP-2026-v1 regime, and release the underlying 860,213-launch record-level dataset under CC-BY-4.0 on Zenodo with a frozen schema at concept DOI 10.5281/zenodo.20633486. Social-channel presence exerts a large effect: launches advertising a Telegram channel graduate at 1.485% versus 0.166% without (8.94x lift, log-rank p 31 SOL) graduates at 0.634%, almost exactly matching the 2025 pooled rate of Marino et al. (2026). The cross-regime decline is substantially attributable to a shift in launch composition toward zero-self-buy tokens rather than to a fall in success rate among self-buy launches. Cox concordance is 0.858.