Abstract: A systematic experimental investigation was conducted on the motion response (RAO) and mooring performance of a novel disk-shaped buoy (geometric scale 1:10) subjected to combined wind, wave, and current actions. A hybrid experimental strategy was employed, integrating a large-scale wave flume (for long-period waves and currents) with a harbor basin (for short-period waves and wind), aiming to mitigate the scale effects inherent in Froude-scaled models, particularly with regard to drag force measurements. The test matrix included free decay in calm water, RAOs under regular waves, motion and mooring line tension under irregular waves, and measurements of wind and current drag coefficients. Key results indicate a natural roll period of approximately 3.0 s with a notably high dimensionless damping ratio (ζ ≈ 0.14–0.15), which is conducive to rapid motion attenuation. A pronounced resonance peak in the roll RAO (26.6°/m) was observed near the 3 s period. Under an extreme sea state (Hₛ = 13.8 m, Tₚ = 16.1 s), the maximum roll angle and dynamic mooring line tension reached 21.30° and 61.56 kN, respectively, the latter being about 3.0 times the static pretension. The mean wind drag coefficient and current drag coefficient were determined as 0.76 and 0.44. This research provides a validated dataset and critical insights for the design, mooring system optimization, and operational safety assessment of such disk-shaped buoys. The effectiveness of the hybrid testing approach is confirmed, and the favorable damping characteristic of this buoy form is highlighted.

Abstract: Agentic AI systems plan over time, call tools, write and retrieve memory, and coordinate across modules and services. Some of the most consequential failures in such systems are structural before they are behavioral: hidden coordination, trace-mediated lock-in, seam bottlenecks, and the silent erosion of meaningful override. This position paper argues that agentic AI systems should be evaluated and governed for \textbf{structural governability}, not only output alignment. By structural governability, we mean whether consequential coordination remains observable, attributable, interruptible, and steerable at the seams between components before irreversible commitments occur. Output-only evaluation does not capture this property. We propose an evidence ladder for structural risk in place of any single master metric: architecture-time priors over system structure, runtime coupling signals on telemetry graphs, and deeper state-regime diagnostics for high-stakes cases. We then sketch a research agenda for benchmarks that stress structure rather than terminal task success, reporting standards that disclose control geometry, and seam-level interventions including approval gates, permission freezes, trace decay, rollback, and subsystem isolation. The wider stake is cognitive integrity. Once agentic systems mediate what users retrieve, remember, delegate, and act upon, alignment depends on preserving the conditions under which users and operators can still understand, contest, redirect, and refuse those processes.

Abstract: Isavuconazole is increasingly being used for the treatment of invasive fungal disease (IFD), although real-world pediatric data remain limited. We retrospectively reviewed patients aged ≤18 years who received isavuconazole for fungal infection at two tertiary centers in Chile between 2021 and 2025. Twenty patients were included, with a median age of 13.7 years (IQR, 9.4–15.5); 14 (70%) had an underlying malignancy, and 6 (30%) were allogeneic hematopoietic cell transplant recipients. Isavuconazole was used for treatment in 16 patients and for prophylaxis in 4, predominantly as second-line therapy due to prior antifungal intolerance, inadequate response, drug–drug interactions, or QT prolongation. Fifteen patients had IFD, with Mucorales (n=3) being the most frequently identified pathogens in proven cases and pulmonary involvement predominating in probable IFD. A succesful response at 90 days was achieved in 60% (6/10) of evaluable cases. No breakthrough fungal infections occurred during treatment or prophylaxis. Hepatic enzyme elevations were observed in four patients. Isavuconazole was associated with favorable outcomes and an acceptable safety profile, supporting its use as an alternative antifungal in pediatric patients.

Abstract: This study examines the relationships between innovation, green management implementation, and their effects on business sustainability and value creation in Indonesian healthcare companies. Innovation is measured using Value Added Intellectual Capital (VAIC) efficiency, while green management implementation is proxied by Environmental, Social, and Governance (ESG) scores. Business sustainability is operationalized through carbon disclosure practices, and value creation is measured via Market Value Added (MVA). Using panel data from 30 listed firms (2019–2023) and applying fixed and random effects regression models showed that green management implementation significantly and positively influences carbon disclosure, supporting the role of ESG practices in enhancing sustainability transparency. However, innovation capacity demonstrates no significant direct effects on either carbon disclosure or market value creation, challenging conventional assumptions about intellectual capital's impact on sustainability and value performance. Component-level analysis improves explanatory power, showing that structural capital efficiency has a marginally negative effect on carbon disclosure, while social ESG scores emerge as the strongest driver of transparency. However, neither carbon disclosure nor ESG performance translates into immediate value creation, with MVA explained more by firm-specific characteristics and profitability (ROA). These findings extend intellectual capital and transparency theories by showing how social performance channels shape disclosure practices.

Abstract: Chalcopyrite and molybdenite exhibit similar surface wettability and high floatability, which has long hindered their efficient and selective separation in mineral processing. In this work, the novel chalcopyrite depressant 2-mercapto-5-benzoimidazole sulfonate dihydrate (2MBI5SA) was investigated for its effect on the flotation behavior of chalcopyrite and molybdenite. Compared with the conventional depressant sodium sulfide (Na₂S), 2MBI5SA exhibited stronger selective depression toward chalcopyrite; under conditions yielding Mo recovery of 81.46% and a Mo grade of 4.46%, the Cu recovery decreased to 13.03%. To clarify the origin of this selectivity, interfacial properties were systematically characterized using adsorption measurements, contact angle measurements, zeta potential measurements, FT-IR, XPS, and SEM-EDS, and the adsorption mechanism was further elucidated using SCC-DFTB calculations. The results demonstrate that 2MBI5SA chemisorbs onto the chalcopyrite surface via bidentate coordination, forming a stable adsorption layer that effectively suppresses chalcopyrite flotation. Moreover, structure-function relationship analysis confirmed that introducing hydrophilic and ionizable functional groups into the collector framework can convert a collector into a selective depressant, thereby providing new insight into the design of environmentally benign flotation depressants.

Abstract: Oncolytic virotherapy offers a promising avenue for solid tumor treatment, yet single-agent approaches are frequently limited by insufficient tumor lysis and inadequate immune activation. Here we report that combined therapy with two recombinant variants of the oncolytic vaccinia virus, armed with either herpes simplex virus thymidine kinase (VV-HSVtk) or the interleukin 15 receptor subunit alpha (VV-mIL15Rα), induces enhanced cytotoxicity and immune stimulation in a murine mammary adenocarcinoma model (4T1). In vitro, VV-HSVtk exhibited dose-dependent cytotoxicity markedly potentiated by ganciclovir (GCV) through HSVtk-mediated phosphorylation into a cytotoxic nucleoside analogue, and co-culture of VV-infected tumor cells with donor-derived NK cells further amplified oncolytic efficiency. In vivo, combined treatment with VV-HSVtk, VV-mIL15Rα, and GCV resulted in significant tumor regression and extended survival relative to monotherapy controls in 4T1 syngeneic mice. Histological examination revealed robust lymphocytic infiltration at tumor sites and absence of hepatic or splenic toxicity. Mechanistically, VV-mIL15Rα-driven IL-15 trans-presentation amplified immune cell activation, while VV-HSVtk/GCV provided targeted tumor debulking and immunogenic cell death, collectively reshaping the immunosuppressive tumor microenvironment. These findings establish a multimodal oncolytic platform combining direct viral cytotoxicity, suicide gene therapy, and cytokine-mediated immune co-stimulation as a translatable strategy for treating solid tumors.

Abstract: Purpose: This study examines whether finance-adapted (FA) phishing detection models improve detection of finance-themed (FT) attacks, whether improvements differ across email and webpage modalities, and whether finance adaptation creates a specialisation–generalisation trade-off. Design/Methodology/Approach: A domain-aware framework is developed using email (164,972 instances) and webpage (11,430 instances) datasets. FT and non-finance-themed (NFT) instances are identified using weighted lexicon-based labelling. Generic models are compared with FA models across Logistic Regression, Linear SVC, and Random Forest using F1-score, MCC, balanced accuracy, ROC-AUC, and PR-AUC. Statistical validation employs bootstrap confidence intervals and McNemar's test, while SHAP and permutation importance interpret webpage model behaviour. Findings: FA models outperform generic models in FT email classification, confirming that finance-specific semantic cues improve detection. However, gains are weaker and less consistent in webpage classification, where models rely mainly on structural indicators (page rank, Google index, hyperlinks). Results reveal a specialisation–generalisation trade-off: FA models improve in-domain detection but do not consistently outperform generic models on NFT instances, with F1-score declines of -0.0057 to -0.0151 on non-finance subsets. Practical Implications: Financial institutions and fintech platforms should deploy domain-adapted detection for email-based threats, where finance-specific linguistic cues yield measurable gains, while maintaining generic or ensemble models for broader webpage phishing coverage. Originality/Value: This study introduces a finance-themed, multi-modal, explainable AI framework for phishing detection, demonstrating that domain adaptation depends critically on data modality and feature representation. It provides a novel systematic comparison of generic versus FA phishing detection across both modalities with statistical validation and explainability analysis.

Abstract: This study proposes a geometric procedure for robust controller tuning under parametric uncertainty, based on root-contour analysis of the closed-loop control system. For a fixed candidate controller tuning, the set of possible pole locations induced by the admissible variations of the control plant parameters is constructed. Robust admissibility is formulated as a geometric set-inclusion problem, requiring this set to remain inside a prescribed dynamic performance region in the complex s-plane. A distinction is introduced between nominal admissibility, robust stability, and robust admissibility, showing that stability over the entire uncertainty set is not sufficient to guarantee the desired dynamic performance. To quantify the root contours, several indices are defined, including the dispersion along the real and imaginary axes, the maximum pole displacement with respect to the nominal pole locations, and the geometric margin to the boundary of the performance region. The procedure is applied to the selection and verification of PI controller tunings for an uncertain single-input single-output (SISO) control system and is further validated through examples with different structures of parametric uncertainty, including a system with a single uncertain parameter and a PID-controlled system with several uncertain control plant parameters. The results show that root-contour analysis can distinguish tunings that are only robustly stable from tunings that preserve the prescribed dynamic performance over the entire uncertainty set. Thus, the method can be used as a practical tool for the diagnosis, comparison, and selection of controller tunings under parametric uncertainty.

Abstract: This paper provides the first systematic, cross-country empirical comparison of the Recovery and Resilience Facility (RRF) and Cohesion Policy funds (CPF) in the domain of renewable energy deployment. Covering 14 EU Member States, the analysis combines quantitative cross-country evidence on financing volumes, technology mixes, implementation speed, and reported capacity achievements. The findings show that the RRF represents a major amplification of EU renewable energy financing, with planned allocations exceeding Cohesion Policy expenditure by a factor of five to ten. At the same time, claims of superior performance-based delivery require qualification: green transition financial progress lags the general RRF disbursement rate, milestone fulfilment for renewable energy falls short of planned indicative rates in most countries, and reported operational capacity figures raise plausibility concerns. The analysis reveals no meaningful correlation between milestone and target fulfilment and progress with renewable energy Country-Specific Recommendations, suggesting that administrative compliance with milestones does not immediately translate into structural reform outcomes. These findings carry direct implications for the design of the post-2027 EU financial framework, particularly regarding the stabilisation of performance indicators, the introduction of attribution protocols for reform-linked achievements, and the preservation of complementarity between performance-based and non-performance-based approaches.

Abstract: Maritime operations rely on the Automatic Identification System (AIS), an open broadcast protocol whose unauthenticated, self-reported Messages are easily abused. This survey takes an AIS-first, security-focused view, grounded in a comprehensive review of prior AIS-security research. We (i) explain how AIS works and use that to expose fundamental weaknesses; (ii) synthesize from the literature the main threats and their technical and operational impacts; (iii) categorize, from the surveyed works and operational practice, mitigations by the layers they target and, for each mitigation, indicate whether it primarily prevents, detects, responds, or supports recovery; and (iv) provide practical recommendations. Bringing together cybersecurity, maritime operations, and data-science perspectives, we consolidate recommendations for securing AIS-based systems and assess their current use in practice, thus highlighting the gaps that standards and implementations still need to address.

Abstract: This study examines the evolution of research on mergers and acquisitions (M&A) in the banking sector through a bibliometric analysis aimed at identifying the main con-tributors, dominant research themes, and emerging gaps in the literature. The study situates banking M&A research within broader discussions of efficiency, market structure, integration performance, and financial stability, while highlighting the need to better understand how scholarly influence and thematic development shape the field over time. Using bibliometric techniques, the analysis evaluates publication trends, journal con-centration, authorship patterns, international collaboration networks, citation struc-tures, keyword co-occurrence, and thematic mapping. The study synthesizes relation-ships among influential publications, institutions, and countries to assess how ideas circulate and which research themes remain central or underexplored within the liter-ature. The findings reveal a concentrated body of scholarship dominated by a limited number of journals, recurrent author networks, and strong transatlantic collaboration patterns, with relatively limited representation from the Global South. Research themes related to efficiency, firm performance, and market structure occupy the core of the literature, whereas technology integration, sustainability considerations, consumer outcomes, and conduct risk remain comparatively peripheral. Citation patterns are highly une-ven, reflecting a small group of highly influential studies alongside a broader set of context-specific contributions. The analysis also identifies a persistent gap between pre-merger strategic narratives and post-merger integration realities, particularly in relation to operational outcomes and systemic risk considerations. The study concludes that future research would benefit from integrating traditional finance approaches with transaction-level integration measures, governance and op-erational performance indicators, and more robust identification strategies around regulatory and policy shocks. The findings further suggest that banking practitioners should place greater emphasis on integration feasibility, risk-control capabilities, digi-tal transformation readiness, and sustainability considerations when evaluating and implementing merger strategies.

Abstract: Infrared object detection from unmanned aerial vehicles (UAVs) is critically challenged by multi-type composite degradation—including noise, blur, and low contrast—which severely undermines feature discriminability and multi-scale target perception. This study proposes SGW-DETR (Spectral-Guided Graph-structured Wavelet Detection Transformer), a novel framework built upon RT-DETR, incorporating three synergistic modules across the backbone, neck, and encoder. FDSANet (Frequency Domain Spectral Awareness Network) replaces the conventional ResNet backbone, integrating the Multi-Scale Frequency Perception Module (MSFPM), Selective Channel Frequency Decomposition (SCFD), and Dynamic Kernel Spectral Modulation (DKSM) to achieve instance-level adaptive spectral feature extraction without degradation-type supervision. The Graph-Structured Fusion Network (GSFN) combines the Adaptive Semantic Fusion Module (ASFM) with the Graph Structure Perception Module (GSPM), employing Gaussian kernel soft membership and two-stage message passing to explicitly model spatial topological dependencies among object components. The Wavelet-guided Contrast Feature Aggregation module (WCFA) restructures the Attention-based Intra-scale Feature Interaction (AIFI) encoder via a Haar-based Frequency Decomposition Unit (HFDU), decomposing features into foreground-edge and background-thermal components and achieving hierarchical foreground–background decoupling through nested dual-path causal contrastive attention. A UAV infrared degradation dataset comprising 4,686 images spanning six degradation types with component-level annotations was constructed for evaluation. SGW-DETR achieves 75.2% mAP50, outperforming RT-DETR by 3.5%, while simultaneously reducing GFLOPs and parameter count by 16.8% and 9.9% at an inference speed of 85.5 FPS. Sustained performance gains on M3FD and IndraEye benchmarks further demonstrate the framework’s cross-domain generalization capability, offering practical value for UAV-based surveillance, search-and-rescue, and border monitoring under adverse imaging conditions.

Abstract: The Gibbs Paradox (concerning the entropy of mixing and entropic extensivity) was explored in depth by Edwin Jaynes (1992). We take up Jaynes’ treatment, considering the special cases for which entropy is (approximately) extensive, and the general case in which it is not. We also explore the Holographic Principle which (strictly speaking) excludes the extensivity of entropy. The formalism of Quantitative Geometrical Thermodynamics shows that, being isomorphic to energy, it is entropy production (not entropy) that is extensive. As a corollary, Shannon information is also not extensive, although information production is extensive.

Abstract: Supernumerary robotic limbs (SRLs) represent an emerging class of wearable robotic systems designed to augment, rather than replace, human motor capabilities. Unlike prostheses or exoskeletons, SRLs operate as independent kinematic agents that enable users to perform multi-limb tasks, reduce physical workload, and enhance operational efficiency in complex environments. This study presents a systematic review of SRL technologies, focusing on mechanical design configurations, sensing modalities, and control strategies, and their influence on key performance metrics such as payload capacity, positioning accuracy, and human–robot interaction efficiency. A structured literature review methodology was adopted following PRISMA guidelines, covering publications from 2010 to 2025 across major scientific databases. The analysis reveals fundamental trade-offs between degrees of freedom, weight, and payload capacity, where high-dexterity systems often impose increased ergonomic burden. Control strategies have evolved from direct teleoperation toward hybrid and shared-autonomy frameworks integrating vision, bio-signals, and machine learning, although challenges remain in achieving intuitive and low-latency interaction. Application domains span industrial manufacturing, construction, rehabilitation, and assistive daily activities, with growing interest in precision-constrained environments such as healthcare. Despite significant progress, limitations persist in actuator back-drivability, long-term wearability, and robust intention recognition under real-world conditions. This review synthesizes current advancements, identifies critical research gaps, and outlines future directions toward scalable, human-centric SRL systems capable of seamless integration into industrial and clinical workflows.

Abstract: Monitoring dynamic post-catastrophic landscapes necessitates unsupervised classification approaches capable of incorporating newly emerging landscape-cover states without relying on predefined classes. Within this framework, the temporal matching of independently derived spectral clusters presents a critical methodological challenge. This study compared alternative temporal matching approaches for multi-temporal Sentinel-2 imagery of the post-catastrophic floodplain landscape of Khortytsia Island (Ukraine) from 2021 to 2026. In addition to conventional methods based on centroid distance, Mahalanobis distance, Linear Discriminant Analysis, and Random Forest, geometrically oriented approaches employing the elongation and principal-axis orientation of spectral point clouds were evaluated. A series of tests assessed matching accuracy, robustness to seasonal and interannual drift, graph connectivity, and consensus structure among alternative matching solutions. The results demonstrated that geometrically oriented approaches preserved temporal correspondence among landscape-cover states with high stability despite phenological and interannual variability. In particular, axis-based matching more effectively maintained separation between corresponding and competing clusters amid progressive temporal divergence. Consensus analysis revealed that disagreement among methods was concentrated in ecotonal and actively transforming zones, indicating areas of increased landscape instability. This study shows that the geometry of spectral trajectories contains valuable information for temporal matching and provides a promising foundation for monitoring dynamic post-catastrophic landscape systems.

Abstract: Accurate information on atmospheric temperature profiles is crucial for improving numerical weather forecasting and short-term numerical weather prediction (NWP). However, the harsh environment of the Tibetan Plateau (TP) limits the availability of station observations, which fails to meet the high spatial resolution required for NWP. In this study, we present a method to calibrate temperature profiles obtained from the Vertical Atmosphere Sounding System (VASS) using data from the polar-orbiting satellite FY-3C. The aim is to provide high-resolution atmospheric structure for NWP in the TP. The temperature profile in VASS exhibits temporal and spatial heterogeneity due to the significant impact of clouds on the radiative transfer mode (RTM). To address this, we employ a combination of variation and artificial neural network (Var-ANN) methods to calibrate the satellite product and improve its compatibility with the model. To confirm the feasibility of our method, we compare the calibrated results with the observed data from 121 radiosonde soundings and 2400 meteorological stations in China, both of which represent conditions closest to the real atmospheric states. The calibrated temperature shows improvements over the original temperature, with a root mean square error, bias, and agreement with radiosonde soundings of 2.11, -0.72, and 0.998, respectively. We also select two classical cases involving the eastward movement of the plateau vortex (PV) and the formation of precipitation to verify the applicability of the calibration in NWP. The results demonstrate that the performance of NWP improves after assimilating the calibrated data, with the Var-ANN data assimilation scheme achieving the highest threat score of 66.9 and 66.7 for case 1 and case 2, respectively. These findings suggest that the Var-ANN method is suitable for calibrating satellite temperature profiles, and the calibrated data holds potential for precipitation forecasting. Furthermore, the novel method can also be applied in global temperature profile correction and satellite cross-calibration.

Abstract: The cross-border Lima River Basin, shared between Portugal and Spain, is prone to recurrent meteorological droughts, which are projected to intensify under climate change. This trend underscores the need for robust early-warning systems to support proactive water management. Under the EU-funded RISC_PLUS project—aimed at strengthening resilience to hydro-climatic risks in the cross-border Minho–Lima River Basins—this study develops a regionalised forecasting framework to evaluate meteorological drought forecast skill using precipitation forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) Seasonal Forecasting System 5 (SEAS5) for the Portuguese section of the Lima River Basin. The 12-month Standardized Precipitation Index (SPI12) is employed as a long-term drought indicator, computed from hybrid 12-month accumulations that combine observed monthly precipitation (October 1979 to February 2025) and SEAS5 forecasts (October 2018 to February 2025). These data are integrated into four hybrid configurations (1 to 6 months lead time) to maximise forecast skill while preserving observed drought memory: 11 months of observations plus 1 month of forecast (11 obs + 1 fcst), 10 obs + 2 fcsts, 9 obs + 3 fcsts, and 6 obs + 6 fcsts. Forecast performance is assessed over the period October 2018 to February 2025. Deterministic SPI12 forecasts and categorical drought classifications are evaluated using a suite of regression-based metrics (e.g., Pearson correlation, root mean square error (RMSE), and skill scores) and contingency-table-based metrics (e.g., false alarm rate (FAR) and F1-score), across SEAS5 ensemble members, percentiles, and spread-based indicators. The 11 obs + 1 fcst configuration, particularly when using the Dry Spread (SpD; defined as the Q10 + Q25 percentiles) and the Q75 percentile, exhibits the highest skill, achieving a Pearson correlation coefficient of r=0.97, an RMSE of approximately 0.17, and near-perfect categorical performance (probability of detection (POD) = 1.00; FAR = 0.00). Conversely, longer lead-time configurations (9 obs + 3 fcsts and 6 obs + 6 fcsts) exhibit degraded performance, with the 6 + 6 configuration providing limited added value relative to climatology. These results demonstrate that SEAS5 precipitation forecasts can provide skilful drought predictions at lead times of up to six months in the Lima River Basin when integrated within the SPI12 framework. The proposed blending methodology therefore provides a robust technical basis for the operational early-warning system being developed under the RISC_PLUS project to support transboundary drought risk management in the Minho–Lima region.

Abstract: Continuous gauge symmetries are usually introduced through Lie groups acting on quantum fields. In this paper we show that the algebraic structure associated with non-abelian gauge symmetry already arises naturally inside the complex group algebra of a finite non-abelian group. The dihedral group D₄, the symmetry group of the square, is used as an explicit example. The complex group algebra C[D₄] decomposes into irreducible matrix blocks under the Artin–Wedderburn theorem. While the character table describes only the subspace of class functions, the full group algebra contains additional intra-class directions invisible to the character table. For D₄ these directions form a three-dimensional subspace which, after elementary normalization, satisfies the Pauli algebra and generates continuous SU(2) transformations inside the two-dimensional irreducible block. The construction is carried out explicitly using only the multiplication table of D₄. The continuity of the complex coefficients allows continuous rotations to arise through exponentials of finite group algebra elements, without requiring the underlying symmetry group itself to be continuous. The mechanism generalizes to any finite group possessing higher-dimensional irreducible representations, where the associated matrix blocks naturally support the corresponding su(N) Lie-algebra structures.

Abstract: Portable biosensor hardware can now sustain continuous multimodal physiological acquisition at the edge, yet the analytical layer that converts raw signals into deployment-consistent inference remains the main bottleneck for practical embedded systems. This study addresses that bottleneck by presenting the machine-learning layer of the Real-time Cognitive Grid, the analytical companion to the previously reported hardware architecture, which equips a fixed-wiring biosensor assembly with real-time physiological-state classification through an asymmetric edge-cloud workflow. The proposed framework assigns analytical responsibility across tiers: a locked 17-feature schema comprising 5 EMG features, 6 EEG spectral features, 2 cross-modal features, 2 HRV features, 1 EOG feature, and 1 EEG quality indicator governs window-bounded inference on the Arduino Nano RP2040 Connect with an LDA edge artefact requiring approximately 716 B RAM, whereas the cloud tier supports public-dataset pretraining, hardware-aligned refinement, multimodal fusion, deployment comparison, and feature-importance analysis under the same schema contract. To evaluate analytical consistency across physiological diversity, five public repositories covering stress physiology (WESAD), affective EEG (DEAP), inertial activity recognition (PAMAP2), sEMG gesture decoding (EMG Gestures), and motor-imagery EEG (EEGMMIDB) were evaluated under subject-disjoint GroupKFold (k=5) protocols. To test whether the same contract survives translation to the physical rig, the hardware branch was evaluated under session-disjoint GroupKFold across five bench-acquired sessions. Unimodal performance was strongest in sEMG- and IMU-dominant tasks, whereas multimodal fusion improved macro-F1 by up to 0.141 over the strongest unimodal baseline in WESAD and by 0.109 in PAMAP2. In the hardware branch, the deployed edge LDA artefact reached 0.9435 macro-F1 with 0.9470 accuracy, while the retained cloud Random Forest reached 0.8792 macro-F1 with 0.8799 accuracy; feature-importance analysis further showed that the final 17-feature branch was dominated by EMG descriptors, with EEG spectral terms contributing secondary support and hardware-exclusive variables remaining weak under the present bench regime. These results show that a compact multimodal sensing assembly can be elevated beyond passive signal capture into an intelligent portable biosensor that performs context-aware interpretation with minimal user intervention, supported by a reproducible analytical workflow that remains coherent across heterogeneous benchmark repositories, hardware-specific refinement, and microcontroller-class deployment, thereby establishing cross-session bench feasibility as a structured basis for future multi-subject wearable validation.

Abstract: This study proposes a novel measurement system repeatability and reproducibility (R&R) framework for zero-inflated correlated defect-count data in semiconductor wafer automated optical inspection (AOI). In advanced semiconductor manufacturing environments, AOI systems are extensively used to detect wafer defects such as particles, scratches, and structural abnormalities. However, conventional Gauge R&R methods are primarily developed for continuous Gaussian-type measurements and are therefore not fully appropriate for high-yield semiconductor inspection data characterized by discrete defect counts, excessive zero observations, and correlated defect categories. To address these limitations, this study develops a zero-inflated bivariate Poisson (ZIBP) measurement system model capable of simultaneously capturing correlated defect-generation mechanisms and structural zero-defect states. A latent-variable representation is introduced to model shared and category-specific defect sources, while a zero-inflation mechanism accounts for defect-free wafer observations commonly encountered in precision manufacturing. An expectation-maximization (EM) algorithm is further developed for parameter estimation, including latent common defect counts and structural-zero probabilities. Based on the fitted model, repeatability variance, reproducibility variance, total measurement variation, and Percent R&R are estimated under the proposed probabilistic framework. In addition, bootstrap resampling is employed to construct confidence intervals for the proposed R&R measures. Theoretical properties of the proposed framework, including covariance structure, identifiability, EM monotonicity, estimator consistency, and asymptotic behavior of the Percent R&R estimator, are analytically established. The proposed framework extends traditional Gauge R&R analysis from continuous Gaussian measurements to zero-inflated correlated count-type defect inspection data and provides a statistically rigorous methodology for evaluating AOI measurement system reliability in semiconductor wafer manufacturing environments.