Abstract: UAV, mounted aerial base stations offer remarkable deployment flexibility for 6G Non Terrestrial Networks (NTNs), yet their hovering energy demands present a critical sustainability bottleneck, particularly as network topologies continuously evolve in practice. As a first step, we develop MARL-ESA (Energy-Saving Adaptation), a Multi Agent Reinforcement Learning (MARL) framework achieving 48% energy savings in static UAV assisted heterogeneous networks (HetNets) compared with always-on operation through centralized training with decentralized execution (CTDE). However, MARL-ESA remains inherently topology specific, requiring over 1500 retraining episodes whenever UAV positions or traffic patterns change. To overcome this limitation, we propose Meta-MARL-ESA, a meta reinforcement learning extension that enables zero shot generalization across unseen network configurations. The problem is formulated as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) over a task distribution τ∼p(T), with a permutation invariant Graph Neural Network (GNN) policy trained via Model Agnostic Meta Learning (MAML). Simulation results demonstrate that Meta-MARL-ESA achieves 12× faster convergence on new topologies, preserves 82% of optimal performance in zero shot deployment, improves energy efficiency (EE) by 21.5% over converged baselines in out of distribution (OOD) scenarios, and recovers to 98% of optimal performance within 8 decision epochs during dynamic UAV repositioning events.

Abstract: Cardiovascular–kidney–metabolic (CKM) syndrome integrates three major chronic disease burdens — heart failure, chronic kidney disease, and metabolic dysfunction — into a unified pathophysiological framework. Accumulating evidence positions biological aging as the central substrate linking CKM risk factors to organ failure. Cardiac aging promotes mitochondrial dysfunction, diastolic stiffening, and impaired repair capacity; renal aging depletes tubular cell renewal, accelerates fibrosis, and reduces the production of the anti-aging hormone α-Klotho; pancreatic β-cell aging drives p16INK4a-mediated senescence and mitochondrial uncoupling that impair first-phase insulin secretion, constituting a primary metabolic driver of early CKM progression. Klotho, a kidney-derived endocrine factor whose circulating levels decline with CKD progression and metabolic disease, regulates cellular proliferation and mitochondrial homeostasis through the Sirt1–CHK2 signaling axis. Three-dimensional electron microscopy of aged renal tubular mitochondria reveals age-dependent loss of cristae structural integrity associated with reduced expression of the MICOS complex—the protein scaffold that maintains cristae junction architecture. Parallel volumetric electron microscopy of human failing myocardium demonstrates distinct mitochondrial structural phenotypes in heart failure with preserved ejection fraction (HFpEF) versus Heart Failure with Reduced Ejection Fraction (HFrEF), underscoring the specificity of mitochondrial remodeling in the CKM context. Measurement of circulating Klotho, senescence-associated secretory phenotype (SASP) markers, and urinary mitochondrial DNA provides a biological staging dimension that complements standard CKM clinical classification. The mitochondria-targeted peptide elamipretide (SS-31) — the most mature pharmacological intervention targeting inner mitochondrial membrane integrity — has demonstrated tremendous functional benefits in numerous pre-clinical models of heart failure, though the effect in HFpEF trial remains inconclusive. Together, these data outline a personalized medicine approach to CKM syndrome in which biological aging phenotype guides therapeutic strategy.

Abstract: Whether a monetary anchor independent of discretionary issuance can remain stable at the macro level is the central open problem identified by the first paper of this program (Cheng and Cheng, 2026d). The present paper takes up that problem under the program’s normative premises and develops a conditional reconstruction of inflation and macrostability. We first specify the complete life cycle of the currency: issuance and extinction are mirror images of the creation and consumption of value. Consumption events trigger the cancellation of the corresponding money at the platform clearing node; durable value is extinguished on an amortization schedule that tracks physical depreciation; money issued against non-rival knowledge persists. The money stock thereby corresponds to the surviving value stock at every instant, M(t) ≡ ks·V(t). On this basis we identify five mechanisms through which inflation is structurally precluded rather than suppressed — temporal synchrony, the quantity identity, channel excision, expectation self-stabilization, and the separation of relative prices from the price level — and we state the structural price-stability proposition with its scope made explicit: the accounting unit, the money price of a unit of negentropic value, satisfies P(t) ≡ ks as a matter of accounting construction, while the consumer price level experienced by households is governed by the dual-price band analyzed in Section 4.7. Inflation is then rediagnosed as the relative deviation produced by four identifiable integrity failures: accounting falsification, consensus distortion, issuance corruption, and extinction failure. The zero-inflation condition, the monotonic structure of consensus distortion, and the existence of a buffering bound for the stability coefficient ks are stated as propositions, together with a viability condition for the metabolism backstop, a composition bound on experienced inflation, an optimal buffering rule under stochastic shocks, and a tipping condition for the transition window in which the new unit coexists with legacy money. The framework’s falsifiable content is discharged computationally: five of the six pre-registered hypotheses are executed in an agent-based stress test (Appendices C and D), yielding three confirmations, one strong confirmation of the extinction layer’s necessity, and one partial falsification with a repair whose scope correction has been folded back into the main text. The sixth hypothesis — a real-world panel test of whether measurement and consensus integrity predict subsequent inflation volatility — cannot run in a model that implements the theory’s own equations; it is therefore specified as a complete pre-analysis plan with named data sources, identification strategy, and decision rules (Appendix E), and executed at baseline depth against that plan (Appendix F), where the pre-registered decision rules return a verdict of not confirmed — correct signs, economically large point estimates, insufficient precision — so that the hypothesis enters the literature with an executable target and an unsoftened first verdict. All simulation parameters are labeled by provenance, and all code is committed to public deposit.

Abstract: Manual annotation remains the source of ground truth for point cloud semantic classification, but its tools assume a contributor who can download billions of points and run desktop software. We present a browser-based framework that lets contributors classify remotely hosted datasets they neither own nor download. Points are labeled with a volumetric brush, and only the octree nodes the brush boundary crosses are fetched: regions fully inside are classified without being loaded, so the cost of a stroke follows its boundary rather than the points it encloses. Edits are held in a small ledger file, separate from the source data, that survives the discarding and reloading of points a streaming viewer performs and can be applied back to the source files at any time. Contributors receive disjoint regions as URLs and work asynchronously, returning ledgers that a project manager merges; classified subsets can also be exported as tiled LAS/LAZ. Class schemas are user-defined and travel with the edits. Where geometry alone is ambiguous, contributors label through pose-registered survey photographs. We demonstrate the framework on cultural heritage stonework, post-fire aerial LiDAR, and coral reef photogrammetry. It is open-source and deployed in production across OpenHeritage3D, the ALERTCalifornia Atlas Archive, and OpenOcean3D.

Abstract: Background/Objectives: Histone H3 K27-altered diffuse midline gliomas (DMGs) have poor outcomes and are characterized by global loss of H3K27 trimethylation (H3K27me3). Evidence from other H3K27me3-deficient tumors suggests a potential vulnerability to combined histone deacetylase (HDAC) and MEK inhibition. We examined a clinical experience with this combination and tested its activity in glioma models. Methods: We reviewed the course of an 11-year-old with neurofibromatosis type 1 and a thoracic spinal DMG treated with selumetinib and vorinostat after gross-total resection, without upfront radiotherapy. In post hoc viability assays, we tested HDAC and MEK inhibition in two H3K27M-mutant glioma lines, an isogenic K27-rescued derivative, and two H3-wildtype lines. Combination benefit was measured as the reduction in area under the viability curve relative to vorinostat alone. Results: The patient experienced 22 months of radiotherapy-free, progression-free survival before recurrence, with a postoperative enhancing focus resolving during treatment. Treatment was associated with grade 1 diarrhea and dermatitis. In vitro, combination benefit was greater in the H3K27M-mutant lines than in the H3-wildtype lines (mean difference, 13.2; 95% confidence interval, 8.1–18.3; p < 0.001). Benefit was also greater in DIPG17 than in its K27-rescued derivative (mean difference, 22.4; 95% confidence interval, 10.1–34.7; p = 0.015). Conclusions: This clinical observation and the post hoc validation experiments support prospective study of H3K27me3 loss as a candidate biomarker for combined HDAC and MEK inhibition.

Abstract: Arterial oxygen saturation measured by co-oximetry (SaO2m) is the clinical reference for oxygenation, but in the mechanically ventilated patients studied here the freshest usable value is nine hours old. We build a virtual sensor that bridges this gap from continuously recorded ventilator and gas-rig time series with gradient-boosted trees, and report it as three nested implementation scenarios of one code base: X0 (XGBoost 2.1.4 through our pipeline, no physics term active), XP (X0 plus a blend toward the anchor and the cumulative oxygen uptake rate over the gap per kilogram of body mass, derived from an oxygen balance of the arterial blood) and v4 (XP plus a patient-level anchor that recalibrates the nine-hour-old measurement toward the patient’s own history). Out-of-fold predictions under patient-grouped five-fold cross-validation are scored on the 3730 hours of 30 patients whose inputs contain no SaO2m value younger than nine hours. From X0 to XP, the mean per-patient MAPE decreases slightly, from 3.321% to 3.289%, an effect resolved by a paired patient bootstrap for this summary only. From XP to v4, it decreases further to 3.117%, the pooled MAPE decreases from 3.890% to 3.727% (both changes resolved by the bootstrap), and the pooled correlation increases from 0.754 to 0.793. Meanwhile, the model becomes substantially smaller, with the number of input features decreasing from 55 to 19 and the number of tree leaves decreasing from 9527 to 450. An exact decomposition of every prediction shows that in v4 the two-coefficient anchor earns 81% of the improvement over persistence and carries nearly all of the prediction variance. All formulas are matched line by line to the C++ and Python sources.

Abstract: Live data migration across geographic regions in Internet-scale cloud services requires distributed locking—but dedicated lock services add infrastructure, latency on every request, and atomicity gaps between lock release and location cutover. We present an alternative: embedding lock metadata directly in the routing records that services already read on every user request. Deployed at a large-scale real-time messaging platform processing 6–10 million migrations per day, this approach makes lock enforcement free for the overwhelming majority of requests—lock checks piggyback on the routing lookup every request already performs—with a single extra point read only for the small, transient set of resources that carry an active lock. It also provides atomic cutover (location swap and lock release in a single database write) and deterministically correct lock propagation (passive cache expiry rather than active invalidation). Write-lock duration—the only user-impacting phase—holds at p99 under 20 seconds. We compare against Chubby, ZooKeeper, Redlock, and DynamoDB Lock Client, and discuss the broader applicability of infrastructure-embedded coordination for Internet-scale distributed systems.

Abstract: In this paper a comparison of labor cost is carried out for apparel manufacturing in different countries. The factors that may have impact on labor cost are analyzed using data spanning from 2001 through 2025. Out of the multiple options for the location of apparel manufacturing and decisions regarding apparel sourcing, lower cost options are generally preferred. However, the landscape has shifted significantly due to rising wages in traditionally low-cost countries, the impact of the COVID-19 pandemic, trade tariff disruptions (particularly US-China trade tensions and 2025 reciprocal tariffs), automation pressures, and sustainability compliance requirements. Analysis is made to estimate the rate of labor cost increase in low wage and high wage countries using updated linear regression models. The global apparel market, valued at approximately $1.77 trillion in 2024, continues to be shaped by these evolving cost dynamics. This revised edition incorporates data through 2025 from the ILO Global Wage Report 2024–25, Werner International, the US Bureau of Labor Statistics, and multiple industry sources.

Abstract: Prior research links wealth transfers to reduced labour force participation, but their effects on intra-household allocation of domestic tasks remain unexplored. This paper explores how inheritances impact gender differences in time use within the household among heterosexual couples older than 50 years in Europe. We use panel data from the Survey of Health, Ageing and Retirement in Europe (SHARE) to analyse the time spent on paid work, unpaid work, and leisure. We employ a Pooled Ordinary Least Squares (OLS) model with fixed effects to estimate the impact of inheritances greater than €5,000 on time allocation to different activities within the household. Specifically, we adopt the intra-household approach to assess ownership, where we consider inheritances as individual assets, rather than the mainstream approach of inheritance pooling that considers inheritances as household assets. Our results show that women significantly reduce their time spent on unpaid work, specially when they are the inheritance owners and when they are aged between 50 and 70 years. Robustness analyses support these results, we find that the mechanism behind the decline in unpaid work is outsourcing to the market, and the results are mainly driven by liquidity constrained households. Our findings suggest that intergenerational wealth transfers in ageing societies shift intra-household dynamics, enhancing women’s bargaining power and reducing inequality in unpaid work among the poorest households.

Abstract: Reliable ginseng authentication requires analytical signatures that distinguish plant category, production region, and growth duration while remaining interpretable and operationally compact. This study evaluated laser-induced breakdown spectroscopy (LIBS) as a proximal elemental-sensing method for these three classification tasks. The datasets comprised 200 spectra from four plant categories, 100 spectra from two geographical origins, and 100 spectra from two growth years; each class contained 50 pressed pellets prepared from one homogenized powder batch. Four chemometric pipelines were compared using five-fold nested cross-validation repeated five times, with all preprocessing and model selection confined to training folds. The best mean balanced accuracies were 1.000, 0.982, and 1.000 for plant category, origin, and growth year, respectively. A sparse model using only 50 of 12,990 wavelengths retained balanced accuracies of 1.000, 0.980, and 1.000. Stable regions were provisionally associated with Ba, Sr, and K for plant category, Al for origin, and Ca and Fe for growth year. Standard normal variate preprocessing suppressed sensitivity to uniform intensity scaling and offsets, whereas detector-channel shifts caused marked deterioration. The results show that compact LIBS fingerprints can encode multiple ginseng attributes and identify wavelength registration, rather than intensity variation alone, as a central requirement for future transfer across batches and measurement settings.

Abstract: Background: Artificial intelligence (AI) is increasingly incorporated into administrative work, yet evidence remains limited on whether specific AI components translate into measurable job redesign and whether organisational culture conditions this relationship. This study examined the relationships between optical character recognition (OCR), chatbots and virtual assistants (CVA), robotic process automation (RPA), organisational culture and job redesign. Objectives: To assess the associations of OCR, CVA and RPA with job redesign and determine whether organisational culture moderates these relationships among administrative personnel. Methods: A cross-sectional quantitative study was conducted among 323 respondents. Descriptive statistics, Pearson correlations, multiple linear regression, and hierarchical moderation regression were used. Statistical inference was based on regression coefficients, F tests, P values and explained variance. Results: Respondents reported relatively favourable levels of OCR (M = 3.882, SD = 0.831), CVA (M = 3.849, SD = 0.873), organisational culture (M = 3.819, SD = 0.849), RPA (M = 3.811, SD = 0.905) and job redesign (M = 3.730, SD = 0.868). Correlations among variables were minimal. Collectively, OCR, CVA, and RPA explained only 0.2% of variance in job redesign, and the regression model was not significant (F(3,319) = 0.165, P = 0.920). None of the predictors was independently significant: OCR (β = 0.014, P = 0.797), CVA (β = 0.021, P = 0.708), and RPA (β = 0.031, P = 0.582). Organisational culture did not significantly moderate the AI–job-redesign relationship (β = 0.018, P = 0.891). Conclusion: AI-component adoption was not associated with measurable job redesign in this sample, and organisational culture did not significantly alter these relationships. AI implementation may therefore enhance specific administrative processes without necessarily restructuring jobs. Recommendation: Institutions should complement AI deployment with task redesign, workflow integration, employee capability development and implementation-readiness strategies rather than assuming that technology adoption alone will transform jobs. Significant statement: These findings challenge a simple technology-deterministic assumption that AI adoption automatically produces job redesign and highlight the importance of examining the organisational and task-level mechanisms through which AI becomes consequential for work.

Abstract: This narrative review examines recent literature on human genetic variation and the clinical spectrum of dengue. A comprehensive search through September 2026 synthesized findings from genetic association studies, HLA typing, and genome-wide studies. The evidence is systematically categorized according to the World Health Organization clinical classification. The majority of available data derive from investigations conducted in Southeast Asia, with limited contributions from Africa and Latin America. Notably, specific HLA alleles—particularly HLA-A*24—as well as MICB and PLCE1 gene variants demonstrate the strongest associations with severe dengue; however, these associations vary by population and viral serotype. Other immune response genes, including CD209, FCGR2A, and TNF, have also been implicated, although their effects exhibit inconsistency across diverse cohorts and epidemiological settings. Conversely, the impact of certain variants, such as FCGR2A R/R131, appears to be modulated by population context, infecting serotype, and infection history. A major limitation of the existing body of research is the prevalence of small sample sizes, heterogeneity in case definitions across guidelines, and insufficient adjustment for genetic diversity. To date, no genetic marker has been validated for routine clinical use on a global scale; nevertheless, large-scale genetic analyses have yielded novel insights that warrant further investigation in large, multi-ethnic studies integrating both host and viral genomic data. Genetic analyses are crucial for improving vaccine efficacy and for identifying therapeutic targets to prevent dengue infection and reduce the risk of severe disease.

Abstract: Microplastics are increasingly recognized as contaminants of emerging concern in drinking water, highlighting the need to assess drinking-water-contact materials as potential sources of particle release. This study investigated microplastic migration from six materials commonly used in drinking water infrastructure under chlorinated and non-chlorinated conditions using three consecutive migration cycles. Particle abundance, polymer composition, and size distribution were determined by micro-Fourier transform infrared spectroscopy (μ-FTIR). Microplastic release was strongly dependent on material composition. Polyester-derived particles predominated in M1 and M2, polyethylene particles in M3 and M4, and PVC particles in M5, whereas the cementitious lining material (M6) exhibited the lowest release potential. Chlorination influenced particle release in a material-dependent manner, with M2 showing the highest concentrations under chlorinated conditions. Across all materials, most particles belonged to the 20-50 μm and 50-100 μm size classes. Consumer exposure was assessed using a Relative Microplastic Release Index (RMRI), identifying M2 and M5 as the major contributors to potential exposure and M4 and M6 as the most stable materials. Overall, the results demonstrate that material composition governs microplastic release and should be considered when evaluating drinking-water-contact materials and consumer exposure.

Abstract: Digital transformation is changing the economic and organizational environment in which financial risks are generated, monitored, and governed. This study examines whether organizational information and analytical capabilities shape financial statement misstatement risk in Indonesian state-owned enterprises (SOEs). Building on the Fraud Hexagon perspective, the study investigates whether pressure, opportunity, rationalization, capability, arrogance, and collusion are associated with financial statement fraud risk (FSF), whether Whistleblowing Systems (WBS) weaken these relationships, and whether Big Data Analytics (BDA) strengthens the mitigating role of WBS. Using 107 firm-year observations from 48 non-financial Indonesian SOEs over 2020–2024 and 25 semi-structured interviews with professionals in auditing, risk management, information technology, cybersecurity, and organizational oversight, the quantitative phase employs variance-based structural equation modeling in WarpPLS 8.0, followed by qualitative thematic analysis. The documented results show that pressure, opportunity, capability, ego/arrogance, and collusion are positively associated with FSF risk, whereas rationalization is not statistically significant. WBS weakens the relationships between the significant Fraud Hexagon dimensions and FSF risk, while BDA strengthens the mitigating association of WBS for pressure, opportunity, capability, ego/arrogance, and collusion. These findings position WBS as an information-generation mechanism and BDA as an information-processing capability within a broader financial risk governance architecture. The study contributes to research on financial risk management, corporate governance, and digital transformation by showing how complementary information capabilities can improve the organizational capacity to identify and assess financial risks in public enterprises. The findings also have implications for risk oversight, resource allocation, internal reporting, and data-enabled governance in digitally transforming SOEs.

Abstract: Ribonucleic acid (RNA) foundation models (FMs) are reshaping computational biology, using self-supervised pretraining on massive transcriptomic corpora to acquire generalisable molecular representations. Developed initially for sequence annotation and expression prediction, they are increasingly applied to plant science, supporting tasks from gene function inference to regulatory element discovery and stress-response modelling in both model and orphan crop species. This review surveys RNA FMs with a focus on plant-relevant applications. We catalogue the major model families, including DNABERT-derived RNA adaptations, PlantRNA-FM, and Nucleotide Transformer variants, comparing their pretraining objectives, input modalities, and evaluation protocols. We examine how these models are deployed across plant genomics, transcriptomics, and functional annotation pipelines, highlighting the opportunities and the constraints arising from species-specific data gaps, limited experimental validation, and interpretability challenges. Across applications, the decisive constraint is no longer architecture but data: corpus breadth, condition-resolved benchmarks, and experimentally anchored validation now dominate the marginal gains available from model design. We close with the key open questions and a roadmap for integrating RNA FMs into the plant systems biology toolkit, emphasising benchmark design, cross-species transfer, and experimentally grounded validation. Throughout, every claim about what a model can do is held to three questions of evidence: what data the model was trained on, which benchmark supports the claim, and whether experimental validation in a plant system exists.

Abstract: Academic inbreeding and recruitment opacity present critical challenges to meritocracy, institutional mobility, and research governance in higher education systems globally. Despite widespread qualitative critiques, quantitative syntheses evaluating the exact degree to which internal candidates or connected candidates are favored remain limited. This systematic review and meta-analysis synthesizes evidence on how institutional ties and recruitment opacity affect academic selection. Following PRISMA-ScR guidelines, searches across Scopus, Web of Science, and PubMed identified 1,742 initial records of which 5 met inclusion criteria. A random-effects meta-analysis using STATA 17 revealed a pooled OR of 3.40 (95% CI: 2.45-4.72), demonstrating that internal candidates or candidates with personal/institutional connections to selection committee members are over three times more likely to be appointed or promoted compared to external applicants. Central, standardized selection committees moderate this advantage (OR= 3.02, 95% CI: 2.25-4.05), whereas decentralized local institutional recruitment exacerbates favoritism (OR= 3.72, 95% CI: 2.81-4.92). ROB 2 domain risk of bias assessments showed overall low-to-moderate bias across included studies. Funnel plot asymmetry indicates minor small-study effects, though Egger's test indicates robust overall estimates. Double-blind initial application screening, mandatory minimum 50% external committee representation, and enforced post-doctoral mobility are necessary to safeguard recruitment integrity and institutional academic output.

Abstract: Pyocyanin is the blue green pigment produce by various strains of Pseudomonas aeruginosa .It has an ability to produce various pigment pyocyanin, pyomelanin, pyoverdine, pyorubin. Bacteria producing Pyocyanin were isolated from the laboratory instruments. CLM11 strain found to produce good amount of pyocyanin and molecular characterization revealed the bacteria is Pseudomonas aeruginosa. Ultra Violet Visible Spectrophotometry and FT-IR confirms the presence of aromatic and phenazine rings and characteristic absorption peaks of pyocyanin. Pyocyanin act as a natural antioxidant by helping to neutralize harmful free radical in the body this will support cellular health and reduce oxidative stress. CLM11pyocyanin had shown significant antioxidant and anti-microbial properties, in vivo toxicity studies in Zebra fish embryos confirms its non-toxic nature and the pigment had shown good dyeing ability to the cloths tested. This study advances creation of an environmentally safe and production process that may be used to replace synthetic dyes and also used as a dye in textile industries.

Abstract: Retrieval of soil organic carbon (SOC) from uncrewed aerial vehicle (UAV) hyperspectral imagery depends on how laboratory measurements are linked to pixels at centimetre resolution. This study presents a quality-controlled workflow in which radiometric calibration, harmonisation against proximal spectroradiometer measurements, pixel-level screening, spectral support definition and model validation are treated as interdependent components of a single procedure. Imagery was acquired with a tunable Fabry–Pérot interferometer camera in six bands between 670 and 1100~nm over eight bare soil fields in Flanders, Belgium (85 topsoil samples) and five fields in Central Macedonia, Greece (42 samples). Single pixel spectra, neighbourhood aggregation and local pixel augmentation were compared at radii of 0.25, 0.50 and 1.00~m using Random Forest, XGBoost, a 1D convolutional neural network and TabPFN, with all pixels from a given sampling location assigned to the same cross-validation fold. In Flanders, 23 of 24 neighbourhood configurations reduced RMSE relative to single pixel extraction, the best reaching \(R^2 = 0.784\) and RMSE =0.155% SOC. In Central Macedonia, adverse acquisition conditions and a narrow target range prevented a usable SOC model, although electrical conductivity, CaCO₃ and pH reached \(R^2\) between 0.61 and 0.69. These results indicate that the spectral support should be defined to match the field sampling support, and that this correspondence belongs alongside radiometric calibration, pixel screening and validation in a reproducible workflow for field-scale soil mapping.

Abstract: Plastic pyrolysis performance depends not only on nominal reactor temperature but also on the actual temperature field established within the reacting bed and vapor phase. This study investigates how the heating-system configuration of a laboratory fixed-bed reactor affects product yields and pyrolysis-oil composition during waste polypropylene pyrolysis. Experiments were performed at 475 °C under N₂ at 0.1 L min⁻¹ for 60 min, comparing a single PID-controlled internal heater with a dual configuration using two independently controlled heating systems. Dual heating increased mean oil yield from 52.14 to 83.95% and reduced gas yield from 45.87 to 14.55%, while solid residue decreased from 1.99 to 1.50%. Temperature profiles showed pronounced overheating in the central vapor-phase region during single heating, whereas dual heating maintained a more homogeneous profile close to the set point. ATR-FTIR, ¹H NMR, and elemental analysis further showed that the dual-heating oil was more aliphatic, with the aliphatic proton fraction increasing from 80.8 to 88.7% and the mean atomic H/C ratio from 2.019 to 2.135. These results demonstrate that heating configuration can substantially alter both yield distribution and liquid-product chemistry even at the same nominal reactor temperature.

Abstract: For targeted backdoor attacks against power-load time-series classification models, attackers can exploit covert temporal triggers to misclassify genuine high-load states as low-load states, leading to issues such as underestimation of load risks and failure of alarms. Since trigger effects may vary across different network modules, existing methods，such as pruning, fine-tuning, and unlearning，typically apply a one-size-fits-all approach, making it difficult to balance backdoor suppression with the accurate identification of normal loads. To the best of our knowledge, this paper is the first to systematically investigate backdoor defense in this scenario, focusing primarily on local time-domain triggers and verifying the method’s applicability under frequency-domain attack conditions. To this end, we propose PRISM-TS, a proxy-trigger-guided personalized backdoor defense method. This method utilizes independent proxy triggers to assess module backdoor sensitivity and jointly selects high-risk modules under clean accuracy constraints. Differentiated cleansing strategies are applied to different modules: high-risk modules are replaced with corresponding modules from the clean reference model, while the remaining modules undergo weighted averaging between the clean and backdoor models. In random time-domain attack experiments on the AEP dataset, PRISM-TS reduced the attack success rate to 2.53% while maintaining a clean accuracy of 95.72% and a high-load retention rate of 89.62%. Frequency-domain extension experiments further demonstrate that this method offers effective defense capabilities.