Social Sciences

Abstract: In coordination games, the distinctive presence of multiple equilibria poses a challenge to coordination, and this has led over time to the proposal of numerous determinants of the latter. In this paper, we study coordination in a repeated stag-hunt game played by two agents who are able to learn from the outcomes of each game, but do not know the payoff matrix. By modulating their learning capabilities, we show that agents are able to coordinate on the social optimum when they learn slowly from experience and are prone to making mistakes. A comparison with the results from a recent laboratory experiment is also provided.

Abstract: Social Laser Theory (SLT) provides a rigorous quantum-like mathematical formalism for modeling collective social activation. While its macroscopic effects mimic laser physics, the fundamental driver of the "lasing" regime resides in the internal state dynamics of the "social atom." In this paper, we propose a quantum-like architecture for the individual cognitive system, formalizing the social atom as a composite quantum system where mental markers—spanning cognitive, affective, and valence dimensions—are represented as internal degrees of freedom in a complex Hilbert space. We show that social activation is not a classical linear response but is governed by a Triple-Resonance Mechanism: the simultaneous quantum-like matching of semantic topic, energetic impact, and valence stance between the agent's internal state and the informational field. We detail how the repeated processing of quantized informational units ("infons") leads to a state of cognitive population inversion. We further establish a link to Human Entanglement Theory (HET), suggesting that the emergence of coherent informational fields is rooted in the phase alignment of these internal mental markers across a population. By situating SLT within the broader framework of Social Quantum Field Theory (SQFT), this paper provides a neuro-cognitive foundation for collective phenomena such as emotional contagion and rapid narrative amplification. Our results suggest that the "Social Laser" is a macroscopic manifestation of quantum-like effects across the human cognitive and neuro-social architecture.

Abstract: The incidence of technologies created with the support of federal funding at universities and federal laboratories that are transferred to the private sector is nowhere close to its potential. The natural question that this observation raises is why has technology transfer research not led to a significantly higher incidence of technology transfer? This paper argues that reification is a primary trait of the analytical frameworks used in technology transfer studies and has detrimental effects on technology transfer research. It examines how reification can adversely impact technology transfer research by distorting our understanding of technology transfer, adversely impacting descriptive research, overstating causal relationships, oversimplifying complex causal mechanisms, and obscuring the role of human agency. It presents a human-centered theoretical framework to mitigate reification. The framework presented has practical implications for technology transfer policy, practice, and research particularly policy design, organizational alignment, and metrics. The suggestions for extending the work presented in this paper focus on empirically testing the HCF and developing measurement instruments to operationalize the human-centered mechanisms of mediation in the framework. This paper contributes to the field by introducing a methodological innovation, identifying a potential flaw in widely used analytical frameworks, and exploring an under-examined topic in technology transfer research.

Abstract: Air transport of lithium-ion batteries has grown faster than the fire-protection systems designed to contain them. The United States Federal Aviation Administration verified a record 89 battery thermal events aboard commercial aircraft in 2024, representing a sixteen percent rise on the previous year, and an independent airline reporting programme recorded a forty percent increase in cargo-side incidents between 2021 and 2025. When a cell fails, it vents for several minutes before producing the visible smoke that current photoelectric detectors are built to catch. That vent gas is not a vague hazard but a chemically defined mixture of hydrogen, carbon monoxide, carbon dioxide and light hydrocarbons, almost all of which falls under Class 2 of the United Nations dangerous goods scheme. This review treats that correspondence as a deliberate design starting point, positioning the Class 2 taxonomy as a sensor architecture input rather than a filing category. The experimental literature on vent gas composition is synthesised and read against the operational record of aviation incidents. An Analytic Hierarchy Process and TOPSIS decision model is then constructed to rank five candidate sensor families—electrochemical (EC), non-dispersive infrared (NDIR), tunable diode laser absorption spectroscopy (TDLAS), metal-oxide semiconductor (MOX) and photoionisation (PID)—against seven criteria covering detection limit, response time, selectivity, flight-envelope tolerance, certification maturity, power draw and cost. The electrochemical sensor ranked first (TOPSIS closeness coefficient C* = 0.741), followed by non-dispersive infrared (C* = 0.635) and tunable diode laser spectroscopy (C* = 0.586). Robustness checks confirmed that the top-three order holds under all twenty-percent single-weight perturbations and across three policy scenarios. Because no single technology covers the full Class 2 envelope, a combined architecture is recommended: an electrochemical hydrogen channel, a non-dispersive infrared channel for the carbon oxides, and a metal-oxide array for hydrocarbon classification and redundancy. This combination aligns with the chemistry-specific findings of an independent principal component analysis of 247 reported failure cases. The review closes with concrete regulatory proposals for the ICAO Technical Instructions, the IATA Dangerous Goods Regulations and the EASA certification basis.

Abstract: Digital twins are becoming vital decision-making infrastructures across critical infrastructure sectors such as smart city urban services, transportation, energy, and healthcare. As digital twins become autonomous and gain real-time intervention capabilities, their governance becomes increasingly essential. Yet existing governance mechanisms remain largely procedural: they emphasize compliance without operationalizing the cognitive practices required to question assumptions, detect algorithmic harms, or support legitimate multi-actor deliberation. Drawing on a systematic scoping review, this study synthesizes the literature on digital twin autonomy, algorithmic risks, epistemic foundations, and governance mechanisms. The review reveals a fundamental gap: current governance mechanisms lack institutionalized cognitive capacities for continuous validation, proactive detection of emerging harms, and structured multi-stakeholder deliberation. To address this gap, the paper proposes the skeptical intelligence framework, developed through design science research. The framework integrates three cognitive functions: validation, detection, and deliberation supported by operational principles, governance artifacts, and distributed accountability roles. The framework advances digital twin governance beyond compliance toward a model rooted in critical epistemology, reflexivity, transparency, and democratic legitimacy. It also lays the groundwork for validation of the proposed framework through implementation across multi-actor digital twin infrastructure contexts—including smart city governance, port logistics, and energy networks—where DT-mediated decisions redistribute opportunities and risks across heterogeneous stakeholders. Empirical validation in an operational setting is planned as the next phase of this research.

Abstract: As the supply network for community-based elderly care services expands, the research focus shifts from mere service availability to computationally modeling who can access and utilize services effectively. Existing studies often consider accessibility as spatial distance or economic cost and equity as simple resource allocation, limiting insights into cumulative disadvantages faced by older adults with low income, digital barriers, or limited family support. Based on data from 3,800 community elder care sites across 20 U.S. metropolitan areas, individual survey data from 6,240 older adults, and community socioeconomic indicators, this study constructs a computational five-stage accessibility chain model: “Information Accessibility—Eligibility Determination—Process Accessibility—Service Availability—Outcome Attainability.”The study integrates heterogeneous data encoding, adaptive spatial accessibility computation, stage-aware vulnerability representation, and hierarchical modeling. Two-Step Floating Contour Analysis (2SFCA), stage-coupled logit modeling, and deep embedded clustering are applied to stratify risk and optimize service access prediction. A cross-vulnerability index, combining six factors—age, income, cognition, language proficiency, family support, and digital access—is incorporated into the model to quantify cumulative impacts across stages.Preliminary results indicate that inequalities do not primarily arise from geographic proximity but accumulate during intermediate phases of information identification, eligibility determination, and process navigation. Digital vulnerability and lack of family support remain major limiting factors even after controlling for spatial accessibility, demonstrating the effectiveness of stage-aware, computationally optimized modeling. This paper proposes an integrated “accessibility chain–intersectional vulnerability” computational framework, advancing service equity from outcome equity to process and transformational equity, and providing a technology-driven foundation for targeted service allocation, navigation system optimization, and identification of high-risk groups in community-based elderly care.

Abstract: Output-based indicators in entrepreneurial ecosystem governance systematically misclassify pre-threshold structural progress as policy failure, because feedback dynamics produce no immediate output signal. This study examines how institutional coordination shapes those dynamics. Using system dynamics modelling, we construct a three-stock model (active startups, entrepreneurial capabilities, and institutional support). Calibration is performed via structured expert elicitation using the Repertory Grid Technique (RGT), enabling institutionally grounded parameter estimation where comparable time-series data are unavailable. Three policy scenarios — fragmented support, financial intensification without coordination, and coordinated early intervention — are simulated for Mexico and the United Kingdom. Resource intensification alone yields only temporary gains when feedback structures remain fragmented. Coordinated intervention activates reinforcing feedback among all three stocks, enabling self-sustaining growth beyond a critical coordination threshold. The United Kingdom crosses this threshold earlier due to stronger baseline conditions; Mexico responds later but with larger proportional gains. The model provides a feedback-structural diagnostic that distinguishes pre-threshold structural assembly from genuine stagnation, with direct implications for the design of evaluation frameworks in fragile institutional contexts. RGT demonstrates potential as a calibration strategy for feedback models in data-sparse settings.

Abstract: A study of elementary counting (of simple clouds of dots by the Munduruku indigenous people of Brazil) is reanalysed in order to compare and contrast three kinds of probability mass functions (PMFs): (i) quantitative response to a discrete range of counts, (ii) the classic Poisson distribution of miscounts, and (iii) psychometric (Rasch) distributions of counting task difficulty and person counting ability. PMFs provide a means of defining — for discrete and qualitative data — the basic metrics, viz. location and dispersion, of metrology — quality-assured measurement, as increasingly required since the turn of the millenium in topical and challenging quality-assurance applications, amongst others, in the human sciences and in Artificial Intelligence. PMF-based metrics, useful in ’clinical’ and other applications where meaning and value are sought, complement the traditionally dominating role played by the corresponding probability density functions (PDF) in ’analytical’, quantitative and continuous Metrology in Physics. New insights are provided when benchmarking the Rasch Poisson Counts Model, which has received less attention in modern metrology, against full psychometric Rasch modelling.

Abstract: Across the Global South, heightened contestation over rural land is placing land administration at the centre of policy attention, as persistent mismatches between official title records and lived realities of occupancy generate legal challenges, political conflicts, and limited access to state programs. Existing systems often alienate landholders who lack valid documentation, limiting their access to welfare and compensation. Digitization of land records is frequently advanced as a solution; however, when implemented without meaningful community inclusion, it risks excluding local voices and producing inequalities in rigid and legally entrenched forms. This article critically examines whether contemporary digitization initiatives adequately address the structural challenges embedded within land administration systems, while also proposing a governance framework that addresses the institutional disconnect between policy design and implementation through decentralization, and co-governance. Drawing on qualitative research from two sites in Western India – Talasari and Chiplun – the study combines Focus Group Discussions (FGDs), field-based Key Informant Interviews (KIIs), and institutional process-mapping conducted between December 2024 and October 2025. The findings show that digitization without community-engaged implementation processes often produces inaccuracies and governance gaps, intensifying fragmentation rather than resolving it, and underscore the need for decentralized, hybrid frameworks that integrate statutory and customary systems through co-governance and community participation.

Abstract: Contemporary large language models (LLMs) generate text by sampling from probability distributions over vocabulary tokens, implicitly asserting that uncertainty about language is well-modeled by calibrated stochastic processes. This paper argues that such probabilistic closure is epistemically unjustified in the regime of language: meaning is bounded by admissible interpretation, not governed by frequentist statistics. The precise claim is not that probability cannot model language but that probabilistic closure is an epistemically overcommitted representation for token uncertainty under bounded admissibility—one that prevents the architecture from representing categorical inadmissibility and compels the assignment of positive probability mass to tokens that should not exist in the support at all. We propose the Possibilistic Language Model (PLM), a novel architecture grounding language generation in possibility theory and the Epistemic Support-Point Filter (ESPF) framework of Jah and Haslett (2025). The PLM replaces softmax probability distributions over tokens with possibilistic compatibility fields over vocabulary support sets; it replaces maximum-likelihood training with a Possibilistic Cramér–Rao regularized entropy-minimization objective; and it replaces standard scaled dot-product attention with Epistemic Possibilistic Attention (EPA), a falsification-driven attention operator that gates keys by admissible innovation geometry rather than by likelihood weighting. The PLM is not merely compatible with these foundations: it is the unique architecture forced by the TEAG axioms when instantiated over a discrete vocabulary manifold. The governing equation of token generation is a tropical Hamilton–Jacobi equation in the max-plus semiring, with the vocabulary surprisal field as Hamiltonian and the PCRB as minimum action per generation step. Epistemic Possibilistic Attention is the discretized Lax–Oleinik operator of this system. The minimax medoid commitment is proved to be the geodesic attractor of the surviving vocabulary well, not a heuristic selection rule. We define (not merely propose) the vocabulary possibility distribution and EPA operator; we derive the PCRB regularization from established ESPF theory inherited from Jah (2026a); we prove, conditionally on VFI maintenance, that the PLM produces non-degenerate generation from actual vocabulary tokens; and we prove that the PLM recovers a standard transformer in the Gaussian epistemic limit, which is the zero-temperature limit of the tropical Hamilton–Jacobi framework established in Jah (2026d). We are explicit throughout about what is fully self-contained here, what is inherited from prior TEAG works, and what remains as open proof obligations. The architecture naturally produces diagnostics—necessity, epistemic width, and surprisal—that quantify when a generation step is epistemically supported versus epistemically strained. We discuss initialization, training, inference, and multi-modal extensions, and identify open theoretical obligations alongside a concrete research agenda. The PLM is not presented as a drop-in replacement for probabilistic LLMs, but as a principled alternative for language tasks where epistemic humility, interpretability, and bounded admissibility matter more than distributional calibration.

Abstract: AI cannot be the property of just five or seven companies in the world because its development and evolution has been the work of hundreds of thousands of researchers and scientists who have been working on it for more than two centuries, some for over two millennia as Pythagoras, Euclid, Aristotle, Al-Khwarizmi and many others. They have left us their legacy in the form of the foundations on which AI stands today. AI is not solely the work of technology company CEO’s, as it has been demonstrated that they have used the intelligence, skills, knowledge and innovations of thousands of anonymous programmers and engineers. It is even less likely to be owned by a government that only cares about its own security and the financial and psychological control of its society. Artificial intelligence is a precious legacy of the work of the most important and valuable foundations on the binary system, originally known as Boolean logic and first described in The Mathematical Analysis of Logic a work published in 1847 by George Boole (1815, 1864) and Formal Logic, written by Augustus De Morgan (1806, 1871), to come together as a tool of incalculable mathematical value in the work of John Venn (1834, 1923) of 1894 in his book Symbolic Logic , from which the concepts for the mathematical treatment of sets and the practical application of the Boolean system were consolidated. Another valuable contribution is the research carried out by Santiago F. Ramón y Cajal (1852, 1934) (Spanish histologist) who obtained important results in his research on The Texture of the Nervous System of Man and Vertebrates (1904), results that were key to the application of artificial intelligence in neural networks. John Bardeen (1908, 1991) andWalter Brattain (1902, 1987) invented the transistor at Bell Laboratories in 1947, based on the theoretical work of Carl Ferdinand Braun (1850 - 1918). The name transistor was coined by John R. Pierce (1910, 2002). Other significant precursors include Gottfried Leibniz, Gottlob Frege, Bertrand Russell and Alfred North Whitehead, David Hilbert, Charles Babbage, John Von Neumann, Claude Shannon, Alan Turing, John McCarthy, Edward Feigenbaum, Douglas Lenat, Judea Pearl, Lotfi Zadeh, John Hopfield and Geoffrey Hinton, as well as hundreds of thousands of unknown engineers. Significant contributions have also been made by research laboratories such as Bell Labs and CERN, as well as thousands of academic research universities around the world. The future of second generation AI will be supported by the work of Thomas Fowler, Jan Lukasiewicz, [1] , [2] Alfred Tarski, Stephen Cole Kleene, the Setun project and scientists, universities and laboratories around the world who are carrying out balanced ternary or fuzzy logic research. The AI must be declared for all the above reasons and more: as part of the Cultural and Technological Heritage of Humanity.

Abstract: Digital inclusion is increasingly recognized as a key driver of socioeconomic opportunity in rapidly urbanizing African cities, yet empirical evidence on its structural determinants remains limited. This study advances the literature by developing a multidimensional, data-driven framework to assess digital inclusion in Ziguinchor, Senegal. Using a unique household survey, it integrates technological access, service quality, affordability, electricity reliability, mobility constraints, and social capital. Principal Component Analysis (PCA) is used to construct standardized domain indices and a composite Digital Inclusion Index, while regression models quantify the relative influence of each domain, accounting for gender and age differences. The findings provide new empirical evidence that digital inclusion is driven primarily by material and infrastructural conditions, particularly device access, proximity and mobility constraints, and electricity reliability. In contrast, affordability and service quality play smaller roles, challenging dominant policy narratives focused on data costs. The study also reveals persistent gender and generational inequalities in digital access and use. By quantifying the relative weight of multidimensional constraints and linking them to spatial and infrastructural conditions, the research offers a replicable and policy-relevant analytical framework for secondary cities. It demonstrates that digital inclusion is not solely a connectivity issue but a structurally embedded outcome, requiring integrated interventions across infrastructure, mobility, and social equity domains.

Abstract: This paper studies platform environments in which participation incentives change discontinuously when valuation crosses salient thresholds. Empirical research on digital and tokenized platforms documents valuation plateaus, abrupt participation shifts, and weak short-run links between price stability and usage dynamics, patterns that are difficult to reconcile with smooth adjustment models. We capture these features structurally by modeling adoption and valuation as a coupled equilibrium system with regime-dependent incentives, formulated as a Filippov differential inclusion.Token prices are determined endogenously through market clearing between usage-driven demand and regime-dependent speculative demand. We establish global well-posedness and identify conditions under which valuation thresholds become attracting manifolds. In these regimes, prices remain anchored at the threshold while adoption continues to evolve, generating persistent valuation plateaus and path dependence. When threshold crossings are regular and the induced dynamics are uniformly dissipative, endogenous boom--bust cycles are ruled out.The framework yields design-relevant insights for digital platforms. Valuation anchoring and volatility depend on governance primitives such as effective circulating supply and regime-dependent speculative depth. A diagnostic numerical analysis links time-at-anchor and sliding intensity to adoption volatility and platform risk, providing empirically interpretable indicators of stability in tokenized platforms.

Abstract: The accuracy of financing demand prediction has a direct impact on the return on investment and risk exposure in fintech investment and asset allocation. Nevertheless, the real world financial transaction data often displays significant nonstationary features — for example, cyclical fluctuations, event shocks, and short-term anomalies — which make the traditional forecasting approach unstable in the real investment scenarios. This study builds a data set that includes 34 reproducible variables — including daily financing requirements, transaction peaks, capital occupation duration, and risk exposure levels — on the basis of 180 consecutive days of investment and operating data from a leading financial services firm. It systematically compares ARIMA, Prophet, Random Forest, and XGBoost models for financing demand forecasting. Empirical results show that XGBoost maintains a low forecast error (MAPE of 8.2%) in the case of market fluctuations and unusual events, which reduces the average error by about 22% in comparison with the baseline model. Based on these results, a model is built to analyze the effect of forecast errors on the stability of investment returns and the efficiency of capital turnover. Results show that keeping the forecast error under 10% significantly reduces the risk of capital misallocation in times of high volatility, while at the same time improving the stability of overall investment returns. This study provides a reusable model workflow and engineering reference for the establishment of the investment allocation and risk management system of the financial institutions.

Abstract: The global regulatory landscape is shifting from voluntary corporate social responsibility (CSR) to mandatory Environmental, Social, and Governance (ESG) disclosure. This study investigates the causal impact of mandatory ESG disclosure on firm value and operational decarbonization using a comprehensive balanced panel of 1,612 listed firms from the EU and the US between 2018 and 2025.Employing a Difference-in-Differences (DiD) design and an event study analysis, our empirical results yield three primary findings. First, consistent with Agency Theory, mandatory disclosure significantly increases firm value (Tobin’s Q) immediately following the 2021 regulatory shock (Post×Treat=0.5212, p< 0.01), indicating that standardized transparency reduces information asymmetry (H1). Second, we document a progressive and cumulative reduction in carbon intensity, providing robust evidence of substantive execution rather than mere ceremonial compliance (H2). The "downward-sloping" trajectory in the event study confirms that the mandate drives real-world operational transitions over time, refuting Decoupling Theory. Third, we find that internal governance mechanisms play a crucial moderating role in this transition (H3); the reduction in carbon intensity is significantly more pronounced in firms with higher board independence and established ESG committees. These findings suggest that "hard-law" transparency mandates effectively align corporate incentives with global climate goals. The synergy between external regulatory pressure and internal governance oversight is essential for bridging the "talk-walk" gap, offering critical implications for global policymakers designing the next generation of climate-related reporting standards.

Abstract: The European Union aims to achieve the target of 25% of land under organic farming by 2030. Italy reached the share of 18.7% in 2022, although significant regional differences persist. This study analyzes farms’ conversion response in the Lazio region (Italy) to evaluate the effectiveness of higher economic incentives in promoting organic conversion. The agro-economic supply model AGRITALIM is applied to a sample of 587 FADN farms. The model simulates individual farm conversion choice, distinguishing between conversion and maintenance phases, and accounting for conversion costs, yield, and price variations associated with each period. Results show limited effects of increased economic support: the 2023–2027 Common Agricultural Policy reform, characterized by higher support, leads to a 5.1% increase in the area under organic farming, while a 40% increase in financial support generates an expansion of 12%. Farm responses are highly heterogeneous: rural provinces, larger and arable farms are more responsive, while smaller farms and livestock are less likely to convert. These findings highlight the need for integrated policy strategies combining financial support, reduced costs, technical assistance, and improved market access. The methodological approach adopted in this study provides a useful tool for supporting the design of targeted and effective policy interventions.

Abstract: Under resource constraints, technology-based SMEs are highly sensitive to the return on training investment. This study analyzes the impact of different training strategies on employee performance, focusing on the relationship between skills gap and training effectiveness. Based on skills assessment, training records, and performance appraisal data of 2,784 technical personnel in a technology-based SME, a skills gap index was constructed, and two types of investment methods were distinguished: general training and job-oriented training. A multiple regression model was used to analyze the relationship between training duration and performance changes. The results show that implementing job-oriented training for employees with skills gaps in the upper quartile resulted in an average performance score improvement of 0.21 standard deviations, while the improvement from general training was less than 0.06. The research results provide a quantitative basis for SMEs to optimize the allocation of training resources.

Abstract: This study aims to design a context-sensitive data governance framework for nonprofit sport organizations in Iran within the era of digital transformation. A qualitative research design grounded in Corbin and Strauss’s grounded theory was employed. Seventeen semi-structured interviews were conducted with managers, coaches, IT specialists, and decision-makers in Iranian nonprofit sport organizations. Participants were selected based on their experience with data-driven projects, decision-making authority, and familiarity with organizational information systems. Data analysis involved open, axial, and selective coding to identify causal, contextual, and intervening conditions influencing data governance. The constant comparative method ensured conceptual consistency and saturation. Key dimensions such as data ownership, quality, infrastructure, organizational culture, and literacy were explored in depth to develop an empirically grounded conceptual model. Results reveal that digital transformation acts as a major causal condition, increasing pressure on organizations to manage large, heterogeneous data sets. Contextual constraints such as limited financial resources, informal structures, and fragmented data processes interact with intervening factors including leadership commitment, staff data literacy, and acceptance of technology to shape governance strategies. These strategies, when implemented, enhance decision quality, transparency, accountability, and organizational trust. The study demonstrates that without formal governance mechanisms, data remain underutilized despite technological adoption.

Abstract: Understanding when and where to shift land from agriculture to forestry is essential for developing sustainable land-use strategies that balance climate, biodiversity, and rural development goals. Traditional profitability comparisons rely on long-term discounting, which is sensitive to assumptions and misaligned with the decision-making horizons of landowners and policymakers. This study introduces a deposit-based framework that treats annual timber biomass growth as accumulating economic value, enabling direct comparison with yearly agricultural profits on a per-hectare basis. By integrating parcel-level spatial data, land quality indicators, national statistics, and expert input, the framework generates high-resolution maps of annual profitability for both land uses. Applied in Latvia, the analysis reveals significant regional variation in agricultural returns, with many low-quality areas showing marginal or negative profits, while forestry offers stable, modest gains across diverse biophysical conditions. The results highlight where afforestation becomes a financially rational alternative and suggest transition pathways that enhance overall land-use profitability while supporting climate and biodiversity objectives. The framework is transferable to other contexts by substituting context-specific data on land quality, prices and growth, and can complement policy instruments such as performance-based CAP payments and afforestation support. The approach supports future-oriented differentiated land-use planning using annually updated spatial economic signals.

Abstract: This article proposes an interpretable, multi-layered recruitment model that balances predictive performance with decision transparency in AI-supported HR processes, ad-dressing risks related to opacity, auditability, and ethically sensitive decision-making. The architecture combines an expert rule layer for minimum-threshold screening, an unsupervised clustering layer to structure candidate profiles and generate pseudo-labels, and a supervised classification layer trained and evaluated via repeated k-fold cross-validation. Model behavior is explained using SHAP to identify feature contribu-tions to cluster assignment, and cluster quality is additionally diagnosed using Necessary Condition Analysis (NCA) to assess minimum competency requirements for attaining a target overall quality level. The approach is illustrated in a Data Scientist recruitment case study, where centroid-based clustering predominates (K-Means is most frequently se-lected), while linear classifiers show the highest effectiveness and stability (logistic re-gression performs best). SHAP highlights competencies that differentiate candidates beyond the initial threshold, and NCA further distinguishes candidates within the recommended cluster by identifying profiles that meet (or fail) the necessary-condition bottleneck. The proposed framework is replicable and supports transparent, auditable recruitment decisions.