Business, Economics and Management

Abstract: Geopolitical conflicts and the energy transition complicate risk linkages among conventional energy, new energy, ESG assets and gold. Using quantile time–frequency connectedness and complex network analysis, this study examines an ESG index, coal and natural gas equity indices, five new energy subsector indices, an aggregate new energy index and a gold ETF across market states and horizons. Under normal conditions, wind power, photovoltaics, natural gas and new energy vehicles are net transmitters, while the remaining markets are net receivers. Total spillovers rise sharply under extreme conditions, and short-term spillovers dominate across market states. Static estimates show stable transmission roles for photovoltaics, wind power, natural gas and gold across quantiles and frequency bands, alongside conditional switching in other markets. Dynamic estimates reveal additional variation over time. The Russia–Ukraine conflict is associated with a sustained increase in long-term connectedness, whereas COVID-19 primarily coincides with short-term volatility pulses. Network analysis identifies coal as a central node and photovoltaics as a major transmission hub within new energy. These findings inform market-state- and horizon-specific risk monitoring, asset allocation and crisis responses.

Abstract: Italy grants registered innovative start-ups fee exemptions, investor tax relief and guaranteed credit, and removes that status sixty months after incorporation. This paper asks whether certification changes firms’ markups. Markups are estimated from a translog production function on 76,466 firm-years of certified start-ups, certified innovative SMEs and ordinary SMEs over 2015–2024. Certified firms price about 12 per cent below ordinary SMEs, but within firms the gap shrinks to 2 per cent and vanishes once persistence is modelled: registration selects firms rather than changing their pricing. Exploiting the statutory expiry, and a 2020 extension that shifted it by one year for adjacent cohorts, the loss of status leaves markups unchanged on average (2SLS −0.005, first-stage F = 750). An unsupervised partition shows this null combines falls of five to six per cent in capital- and materials-intensive firms with none elsewhere; regression learners and double machine learning confirm the central estimates.

Abstract: Studies of small-firm profitability estimate financial structure as the focal regressor and treat cost composition as an unexamined control. This paper reverses the emphasis and places both blocks in one equation, estimated on 91,756 firm-year observations covering 14,913 unlisted Italian firms across three regulatory regimes. Three methods interrogate different assumptions on the same sample: panel estimation across fourteen specifications, unsupervised partitioning selected on eleven validity criteria, and machine-learning regression applied to the same firm-demeaned data. The capitalisation coefficient is not identified. It moves from +0.177 under two-way fixed effects to between −0.12 and −0.28 under five instrument sets, all of which fail because profitability persists at 0.271, so any lagged financial ratio embeds past profit. The labour-share coefficient is stable across estimators, groups and functional forms, and the paper tests whether that stability is arithmetic, since return on assets and the labour share are consecutive lines of one statement. It is not: value added over total assets varies by a factor of 2.8 across the four operating archetypes recovered from the data, while the coefficient varies by 1.07. The accounting identity explains part of the magnitude and none of the stability.

Abstract: This paper investigates the macroeconomic determinants of renewable energy consumption in Saudi Arabia during 1990–2025 using a hybrid approach of ARDL bounds testing and Random Forest machine learning with SHAP (SHapley Additive ExPlanations) to improve model interpretability. We employ annual data from the World Development Indicators to examine the effect of high-technology exports, trade openness, foreign direct investment, inflation and GDP on renewable energy consumption. Our results indicate that renewable energy consumption is highly path dependent, with previous adoption playing a significant role in current consumption (β = 0.776, p < 0.01). Trade openness also shows a negative contemporaneous effect ( = 0.0013, p 0.05) in line with the carbon lock-in hypothesis that typifies hydrocarbon-dependent economies. High-technology exports as a proxy for technological innovation do not appear to be a significant driver, indicating that the innovation-led energy transition in Saudi Arabia is still maturing. The Random Forest model confirms the importance of persistence effects explaining about 73 % of the variance. The analysis of the different time periods indicates that, after 2016 (also known as Vision 2030), there has been a significant change in the structure of the analyzed data, with the statistical model of research becoming much stronger as its explanatory power has reached up to 83%. The analysis of the SHAP allows understanding of the obtained results by measuring the influence of each individual variable. The findings that were obtained pertain to Sustainable Development Goal 7 (which focuses on developing cheap and clean energy), Sustainable Development Goal 9 (which focuses on industry, innovation, and infrastructure), and Sustainable Development Goal 13 (which focuses on climate action). It is our proposal that Saudi Arabia adopt focused innovation strategies, careful management of its trade integration, and a commitment to the institutional framework in order to accelerate the transition to renewable energy. However, it is important to note that this recommendation is suggestive and not definitive.

Abstract: Water stress is read in the sustainability literature as either a physical or an institutional outcome, but the two readings have never been tested jointly within the sovereign ESG framework. This paper estimates the environmental, social and governance determinants of the freshwater withdrawal-to-availability ratio across up to 170 economies observed between 2002 and 2022, drawing on the World Bank Sovereign ESG Data Portal. Three equations share the same dependent variable and differ only in the pillar supplying the regressors. Each is developed along three tracks: panel econometrics, comprising pooled OLS, fixed and random effects, between and weighted least squares estimators and two-step difference and system GMM; a competition among six clustering algorithms assessed on eleven internal validity indices; and a competition among six supervised learners used to validate the specification rather than to forecast. Water stress proves overwhelmingly structural, with 99.1 per cent of its log variance lying between countries. The governance block explains 38 per cent of cross-sectional and 4 per cent of within-country variation. The level of institutional quality, measured as the first principal component of four Worldwide Governance Indicators and absorbing 91.9 per cent of their variance, is unrelated to water stress in every estimator; its composition is not, with accountability relative to state capacity carrying a coefficient of −1.10 between countries. The dominant predictor across all three tracks is the female-to-male labour force participation ratio, whose partial dependence reveals a discontinuity concentrated in a five-point interval. Under cross-validation with countries held out, no flexible learner outperforms ordinary least squares.

Abstract: Distance from equilibrium is treated as a scalar for strategic systems. A local reading, theasymmetry of the equilibrium response to payoff perturbations, and a global reading, theentropy production of the revision dynamics, are used interchangeably because both vanishon potential games. This paper measures whether they carry the same information. Onfamilies of two-player games generated at a prescribed harmonic fraction α of the Candogandecomposition, the two readings co-move almost perfectly across the family (SpearmanρS = 0.9927 [0.9914,0.9936], n= 1000), and that agreement is a confound: each reading isseparately monotone in α (0.9904 and 0.9819). Conditioning on α removes it. Within a levelthe association falls from +0.851 at α= 0.05, at m= 3 and n= 200, to an upper 95% intervalendpoint below 0.35 at every action count tested, m∈{3,4,5,6}, worst +0.156, with thelow-α and high-α intervals disjoint at every m. The two readings share a zero and little else,so an observed system should be located by a point in a plane and not by a number on a line.No system measured here carries both coordinates, so the plane is demonstrated on syntheticfamilies and anchored, not populated, by the two real readings. The structural possibilitythat a local response object and a global dissipation object come apart is not new and is notclaimed here: it is established for Markov jump processes and is exactly false in the Langevinclass, which makes it class-dependent. What is reported is the instantiation of both readingson finite strategic-form games, the harmonic fraction as the computable coordinate alongwhich the breakdown is parameterised, and the measurement. The evidence is two-playerthroughout. At three and four players the low-α coupling that the stratified design needs asa baseline falls below the registered 0.55 precondition (+0.378 and +0.172, against +0.847at N = 2), so the registered player-count kill-shot returns indeterminate instead of a pass,and the finite-size kill-shot returns indeterminate under the registration that binds it. Bothverdicts, a criterion substitution inside the finite-size unit, and a refuted repair are reportedin the body.

Abstract: Logit revision dynamics on a finite game form a Markov jump process on the joint actionprofile, and moving the precision parameter λ drives that process out of its stationary state.This paper measures what the move costs. On the Hatano–Sasa split of entropy productioninto a housekeeping part and an excess part, instantiated exactly on the joint-profile Glaubergenerator, the cost structure inverts along the potential-to-harmonic axis of the Hodgedecomposition. Potential games have identically zero housekeeping, pay only for the change,and obey a discrete quasi-static law ⟨Y⟩∝1/K in the number of protocol steps, withmeasured exponents−1.02 and−0.94: they can be driven arbitrarily cheaply. Near-harmonicgames pay a rent per unit time whether or not they are driven, accumulating 12.4 nats ofhousekeeping at a burn rate of 0.77 nats per unit time at harmonic fraction α= 0.95, whiletheir excess collapses three decades across the same family, from 3.6 ×10⁻² to 2.8 ×10⁻⁵nats. The mechanism behind the collapse was identified by an information-gain-driven probecampaign as a path-length floor over the sequence of steady states, at posterior 0.996 onprobes lying in the well-relaxed regime, where the winning hypothesis is exactly true byconstruction, so that the posterior identifies a regime and is not an empirical discriminationbetween four live possibilities; the floor carries a held-out median residual of 0.002 dexon nineteen unconsumed probes, and a per-step path-aware recursion predicts the excessto a held-out median of 0.076 dex and a worst case of 0.44 dex. The second result is acost of measurement: because the relaxation time sets the hold length a data-side readingrequires, and the spectral gap closes as precision concentrates, the data a real system mustsupply grows precisely in the regime of interest. That floor is quantified with a physicalgate at four relaxation times per hold, a twenty-seed unbiasedness and coverage study, anda decomposition showing that interval coverage survives to n = 20 trajectories while theinstrument’s own verdict does not, failing at n= 30 for every candidate error bar becausethe relaxation-time estimate carries a 35 to 40 per cent across-seed standard deviation atthat same n= 30, over twenty seeds. The empirical position is a refutation reported as such:applied to a real power market, the repeated-protocol premise the whole construction restson fails month by month, with five of seven months anomalous under seasonal drift, and theclosed daily-loop affinity of about 7.0 nats per day [6.22,8.50] stands as descriptive only, onout-of-scope states and with the fluctuation theorem uninformative on cycles. Every othervalue quoted above is a point recorded without an interval, against a project rule requiringone on each quantitative statement: the two intervals given attach to a fluctuation-theoremcheck that is a regression test and to a number reported as descriptive only.

Abstract: Against the backdrop of frequent external shocks and intertwined agricultural production risks, enhancing the resilience of the agricultural industry chain has become crucial for ensuring food security and promoting sustainable agricultural development. As a key national grain-producing province, Anhui Province’s level of agricultural industry chain resilience directly impacts the stability of agricultural product supply at both the regional and national levels. To elucidate the transmission pathways through which the digital technologies influence the resilience of agricultural industrial chains and to investigate the regional heterogeneity of the digital technology’s enabling effects, this study utilizes the panel data from 16 prefecture-level cities in Anhui Province spanning the period 2015-2024 as a sample, this study constructs an indicator system for agricultural industry chain resilience and an indicator system for digital technologies. By employing a two-way fixed-effects model, a mediation effect model, and a panel threshold regression model, this study empirically examines the direct impact of digital technologies on the resilience of Anhui’s agricultural industry chain, the mediating transmission mechanism through agricultural technological progress, regional heterogeneity characteristics, and the threshold effect based on agricultural production risks. The study reveals the following findings: First, the digital technologies exert a significant and robust positive influence on the resilience of agricultural industrial chains in Anhui Province. Second, the agricultural technological progress serves as a key mediating pathway through which the digital technologies enhance the resilience. Third, the impact of digital technologies exhibits pronounced regional heterogeneity. Fourth, the effect of digital technologies on agricultural industrial chain resilience demonstrates a single-threshold effect based on agricultural production risks. Accordingly, this study proposes the targeted policy implications, including further embedding the digital technology into the entire agricultural industrial chain, strengthening the mediating driving effect of agricultural technological progress, implementing the differentiated digital agricultural policies across regions, and prioritizing the grain yield improvement in low-productivity areas. This research provides the solid theoretical and empirical evidence for Anhui and other major agricultural provinces to promote the high-quality digital rural development.

Abstract: strataqis an open-source Python library, built on JAX, for logit quantal response equilibriumanalysis and for four capabilities that are routinely wanted after an equilibrium and thata dated prior-art sweep found unpackaged: the exact derivative of the equilibrium withrespect to payoffs, available in closed form as the strategic resolvent χ^eq = (I−SB)⁻¹S onthe tangent space; the potential/harmonic decomposition of a finite game and its harmonicfraction α; the entropy production and stationary currents of the joint-profile revision chain,together with trajectory estimators calibrated against that exact meter; and a panel-awareestimation workflow whose interval names its own method.The paper states what each function computes, then reports the evidence that eachreading is correct on systems whose answer is fixed in advance by an analytic identity or byan external implementation. Nineteen calibration rows are reported: fifteen carry a declaredtolerance, fourteen a sample size and seventeen a committed artifact, and every exceptionis named in the tables. Most margins are wide because most rows are machine-precisionidentities, where the tolerance is a floor against catastrophic error. The row whose declaredtolerance actually constrains is the agreement with pygambit’s homotopy, at 1.2 ×10⁻⁹against a declared 10⁻⁸. The release-integrity failure in which twenty-three in-repositorygates were green while every solver call in the shipped wheel raised FileNotFoundError isreported as a failure mode with its diagnostic and its mitigation.

Abstract: This study articulates, documents and characterizes a processed financial dataset of daily coffee, Brent oil and gold futures prices for the period 2016-2025. The database was constructed from historical closing prices obtained from Investing.com and organized in long format, so that each record corresponds to a valid daily observation of a specific commodity. Using the original series, a reproducible transformation chain was developed, including logarithmic prices, logarithmic returns, absolute returns, squared returns, 30-day annualized rolling volatility, standardized returns, base-100 price indices, temporal labels and quality-control fields. This manuscript specifically documents the dataset architecture, the information source, quotation units, data-cleaning criteria, temporal consistency and analytical reuse possibilities. The empirical characterization shows that the three markets preserve financial features such as non-Gaussian distributions, heavy tails, heterogeneous volatility and more visible dependence in additional second-moment transformations. The resource is designed to be reused in financial econometrics, econophysics, commodity risk analysis, forecasting exercises, Monte Carlo simulation, financial time-series teaching and comparative studies of agricultural, energy and precious-metal markets.

Abstract: Geopolitical uncertainty has become an increasingly relevant source of information for financial markets, yet whether it improves the out-of-sample forecasting of emerging-market stock returns remains unclear. This study examines whether geopolitical risk (GPR), together with global financial information, improves forecasts of Thai stock-market returns. Using monthly data from January 1990 to July 2026, the study compares ARIMA–GARCH and ARIMAX–GARCH benchmarks with Random Forest, XGBoost, and LightGBM across 1-, 3-, 6-, and 12-month forecasting horizons. An expanding-window rolling-origin framework is employed, with forecast accuracy evaluated using RMSE and MAE and statistical differences assessed using the Diebold–Mariano test and Model Confidence Set procedure. SHAP analysis is used to interpret predictor contributions. The results show that machine-learning models substantially outperform the econometric benchmarks, with LightGBM providing the most consistent forecasting performance across horizons. GPR contributes additional predictive information at the 1- and 3-month horizons, although its contribution weakens at longer horizons. GPR also emerges as the most influential predictor in the LightGBM model. The findings highlight the value of integrating geopolitical information into nonlinear forecasting frameworks for Thai stock-market returns.

Abstract: Agribusiness groups need to connect fast-moving markets, long production cycles, operating indicators, and financial results. In practice, these data often sit in different systems and use different periods, units, and organizational definitions. This article develops a finance-led digital intelligence framework from the experience of building a platform for a multi-unit Chinese agribusiness organization. The analysis identifies breaks between data, calculations, screens, review, and day-to-day operation. It then compares three record structures with four fixed examples and five clear design criteria. The resulting framework has five connected elements: a clear management question, a common data language, interpretable analysis, managerial review, and continuous operation. At its center is a \emph{governed analytical value}. Put simply, the system records a calculated value, each time that value is released, and each review of that release as separate but linked records. This keeps the meaning, accounting basis, method, access rights, release status, and review history traceable. A legacy financial screen borrowed only the indicator-grouping and risk-level ideas of Wei and Wang (2024). It used a separate fixed-rule screen and did not reproduce that study's factor-analysis and logistic-regression model. The article contributes a practical design framework and three propositions that later studies can test.

Abstract: European regions aim to achieve climate neutrality. This requires the improvement of the carbon efficiency of their energy systems despite the substantial differences in climate conditions and economic structure. This study investigates whether carbon intensity converges across European NUTS-2 regions and examines the role of climate-driven energy demand and spatial interactions in the regional energy transition. Specifically, we use panel data from the period 2000–2024 and investigate the conditional β-convergence based on two-way fixed effects and spatial panel models. The results reveal significant convergence, persistent spatial dependence and slower adjustment in regions with initially high carbon intensity. Climate-driven energy demand, labor market conditions and industrial structure significantly affect the convergence process. These findings highlight the importance of place-based energy transition policies that account for regional heterogeneity and spatial spillovers to accelerate decarbonization.

Abstract: This study develops an integrated Explain–Predict–Classify framework to investigate the determinants, territorial heterogeneity, and predictability of regional innovation performance. Using Regional Innovation Scoreboard data for 246 regional units across 31 European countries over 2016–2023, the analysis combines panel-data econometrics, unsupervised clustering, and supervised machine-learning regression. The Summary Innovation Index (SII) is examined alongside indicators capturing scientific collaboration, non-R&D innovation expenditure, SME product and process innovation, collaborative networks, design applications, innovative sales, and environmental conditions. Econometric results show positive and statistically significant associations across all selected innovation dimensions. The Hausman test favors fixed effects over random effects, while the dynamic specification indicates significant persistence in regional innovation performance, although instrument-validity diagnostics require caution. The clustering analysis compares six algorithms using multiple internal validation criteria. K-Means provides the strongest overall solution, with R² = 0.6301, a Calinski–Harabasz index of 370.60, and relatively balanced cluster sizes, revealing ten heterogeneous and partially overlapping regional innovation profiles. The predictive analysis compares seven regression algorithms. K-Nearest Neighbors achieves the strongest test performance (R² = 0.9482; RMSE = 7.90; MAE = 4.793), followed by Random Forest (R² = 0.9289). Permutation importance identifies international scientific co-publications, design applications, and SME collaboration as the most influential KNN predictors. Overall, the findings demonstrate that regional innovation combines common systematic relationships with heterogeneous territorial configurations and nonlinear predictive structures. Integrating econometrics, clustering, and machine learning therefore provides a richer empirical basis for understanding regional innovation and designing differentiated, place-sensitive innovation policies.

Abstract: Between 28 February and 31 March 2026, Brent crude went from $72.48 to a peak of $118.35 per barrel, the Dutch Title Transfer Facility (TTF) gas benchmark doubled, and the Asian Japan Korea Marker (JKM) spot price for liquefied natural gas (LNG) rose by more than 140%. The trigger was the war against Iran and the closure of the Strait of Hormuz, through which about one fifth of the world's oil and LNG used to pass. This article aims to explain a consequence of that shock which the energy literature and the public-debt literature have each studied on their own but almost never together: supply-driven inflation quietly erodes the real value of sovereign liabilities. Three days before the first airstrikes, the Institute of International Finance reported a record $348 trillion in global debt. We trace the transmission channels of the 2026 shock (LNG supply, shipping and war-risk insurance, fertilizers and food), review the dilemma it created for the European Central Bank (ECB) and the Federal Reserve, and compute a simple liquidation arithmetic: each percentage point of unanticipated inflation transfers roughly $407 billion from bondholders to the United States Treasury alone. The article argues that the observed tolerance of 3-5% inflation in 2026 is best read as the initial phase of a financial-repression regime, and that this liquidation channel is a privilege of reserve-currency issuers, while energy-importing emerging economies suffer the opposite effect.

Abstract: This paper examines the dynamic effects of economic policy uncertaintyon key macroeconomic outcomes in Sweden—industrial production, inflation, and unemployment—over the period 2014M01–2025M04. Using the Swedish News-Based Economic Policy Uncertainty Index and applying the Local Projections framework of Jordà, we trace the impulse responses of each variable to a one-standard-deviation EPU shock across a 12-month horizon. Our findings reveal a distinct temporal transmission pattern: industrial production contracts immediately and significantly, with the largest decline occurring around the second month following the shock. In contrast, inflation responds with a notable delay, exhibiting insignificant short-run effects but turning significantly negative from the eighth month onward, consistent with demand-driven disinflation. Unemployment rises more gradually but persistently, becoming statistically significant from the second month and remaining elevated throughout the horizon. These results underscore the importance of a dynamic, horizon-specific approach to understanding uncertainty shocks in a small open economy. From a policy standpoint, the findings highlight the value of forward-looking communication, automatic stabilizers, and targeted labor-market interventions to mitigate short-run output losses and persistent employment deterioration.

Abstract: This study provides a comparative econometric analysis of the impact of artifi-cial intelligence (AI) and digitalization on economic growth in Saudi Arabia over the period 2000–2024. Due to the absence of direct AI investment data, proxy variables including internet usage, government expenditure, inflation, and patent applications are employed. The empirical framework integrates Ordinary Least Squares (OLS), ARIMAX, and ARDL/ECM models to distinguish between short-run dynamics and long-run relationships. The results indicate that short-run effects are generally weak, with only one variable showing marginal statistical significance and a transitional effect. Comparative analysis reveals that although ARIMAX demonstrates better in-sample fit, it suffers from diagnostic limitations and high forecasting errors. In contrast, the ARDL model emerges as the most appropriate framework for capturing long-run equilibrium relationships. The findings suggest that the economic impact of AI and digitalization is primarily long-term rather than immediate. This study contributes by providing a structured comparison between econo-metric approaches and offers policy-relevant insights aligned with Saudi Vision 2030.

Abstract: Logistics infrastructure plays a strategic role in supply chain integration, regional connectivity, and the territorial organization of freight flows in emerging economies. However, the socioeconomic performance of logistics complexes is not automatic or homogeneous, as it depends on institutional capacity, linkages with local business networks, public investment, logistics services, and the development conditions of each region. This study proposes, applies, and validates a transferable composite index to assess the territorial and socioeconomic performance of nine logistics complexes in Colombia. The methodology compares a baseline prior to the operation or consolidation of each complex with a five-year follow-up observation, integrating economic, business, fiscal, social, labor, and distributive variables. The variables were transformed to a base of 100, weighted using principal component analysis, and assessed through adequacy tests, sensitivity analysis, entropy weighting, Spearman rank correlation, and Winsorization. The results show relative improvements across all cases, although with substantial differences in magnitude and internal composition. The Caribbean Logistics Complex and the Bogotá Free Trade Zone obtain the highest index values, while the remaining complexes show positive but more moderate progress. The main contribution of this study is to provide a decision-support tool for logistics planning, benchmarking, infrastructure prioritization, and territorial performance assessment, particularly in emerging economies where logistics investment decisions require the integration of economic development, supply chain connectivity, public investment, employment, social investment, and inequality-reduction criteria.

Abstract: The relationship between public education spending and economic growth remains contested in developing economies, where quality, not just quantity, of investment may matter most. This study examines whether public education expenditure has driven economic growth in Egypt over 1980-2023, addressing the under-studied role of edu-cation quality and structural breaks linked to major reforms. Using the Autoregressive Distributed Lag (ARDL) bounds testing approach with Zivot-Andrews structural break and CUSUM/CUSUMSQ stability tests, the study models short- and long-run effects of education expenditure on GDP growth, controlling for capital formation, labor par-ticipation, trade openness, and foreign investment, and incorporating quality proxies such as student-teacher ratios and completion rates. The bound F-statistic (6.23) con-firm long-run cointegration, and the error-correction coefficient (-0.835, p< 0.01) indi-cates rapid adjustment to equilibrium. Physical capital is the strongest driver of long-run growth, while education expenditure shows a weakly negative long-run as-sociation (significant at 10%), suggesting allocative inefficiency rather than absent re-turns; quality proxies show significant adjustment effects. Stability tests confirm the long-run relationship is structurally invariant once short-run dynamics are accounted for. Improving the efficiency and quality of education spending, rather than its volume, thus appears essential for Egypt to advance Vision 2030, SDG 4, and SDG 8.

Abstract: Electricity-generation mixes can converge in a final-year snapshot while following different transformation pathways, which matters for identifying energy-security exposures. This article analyzes EU-27 electricity-generation mixes for 2000–2024 (27 countries, 25 years, eight generation components) as compositional time trajectories. Shares were regularized by simple multiplicative replacement, transformed with the isometric log-ratio transformation, compared using multivariate dynamic time warping with a Sakoe–Chiba constraint, and clustered with partitioning around medoids (PAM; k-medoids). The selected four-cluster solution is represented by Spain, Bulgaria, Malta, and Latvia and describes broad renewable diversification, solid-fossil-fuel legacy transformation, high-concentration island transition, and renewable-dominant restructuring. The raw-share baseline and the compositional trajectory solution agree weakly (Adjusted Rand Index = 0.1891), while endpoint-only clustering for 2024 also diverges from full-trajectory clustering (Adjusted Rand Index = 0.1950; 10 countries change assignment after label alignment). Across 40 robustness specifications, 14 countries are core cases and 13 are border cases. This boundary sensitivity is a substantive diagnostic: stable interpretation of border countries requires information beyond generation shares. Final mix similarity is therefore not a substitute for pathway similarity, and structural energy-security exposure should be interpreted through diversification, concentration, and the sequence of fuel substitution.