Li Weigang

Purpose: This study aims to discuss the state of the art in Lean Office, its technological advancement through artificial intelligence (AI) and its application in the public sector, while proposing a way to unify these three areas. Design/methodology/approach: A systematic literature review (SLR) following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses protocol identified literature published between 2020 and 2024. The final sample underwent bibliometric analysis (Bibliometrix and R language) to map the field’s conceptual and intellectual structures. Findings: “Lean Principles” and “Public Sector” emerged as motor themes alongside “AI,” with “Lean Office” positioned as a transitional topic. The intellectual structure centers on Lean public sector adaptation. The analysis identifies two integration modes: AI for automating administrative tasks and AI as a support tool for Lean methods. However, tension exists between centralized AI development and the decentralized nature of Lean culture and principles. Research limitations/implications: Despite querying multiple databases, literature simultaneously addressing AI, Lean Office and Public Administration is scarce, indicating an underexplored niche. Additionally, software functionality constraints limited the bibliometric analysis. These findings highlight a knowledge gap and the pressing need for further extensive literature reviews. Practical implications: This study provides information on current research, trends and analysis regarding AI and Lean Office for the public administration community. Originality/value: This research bridges the knowledge gap by integrating three disparate themes: Lean Office, AI and public administration. The findings offer a novel conceptual lens, enabling stakeholders to leverage synergistic relationships between these advancements and helping practitioners interpret complex concepts.

Accurate and transferable land-use and land-cover (LULC) mapping in the Brazilian Cerrado is challenged by spatial and temporal heterogeneity, domain shift, and the high cost of expert annotation. This systematic review with qualitative methodological synthesizes evidence on Active Learning (AL), domain adaptation, remote-sensing foundation models, and Remote-Sensing Vision–Language Models (RS-VLMs), focusing on their complementary roles in annotation-efficient and transferable LULC mapping. Evidence was synthesized qualitatively because the reviewed studies differed substantially in sensors, tasks, datasets, domain-shift conditions, and evaluation protocols. The literature indicates that uncertainty-, diversity-, and representativeness-based AL can improve annotation efficiency, while foundation models and RS-VLMs provide increasingly transferable representations and few-shot or open-vocabulary capabilities across remote-sensing applications. However, direct empirical evidence integrating transferable representations, AL, and domain adaptation for Cerrado LULC monitoring remains limited. Based on the convergence of these evidence streams, we propose a conceptual human-in-the-loop framework that combines transferable representations with selective target-domain annotation and subsequent adaptation. The framework is hypothesis-driven rather than experimentally validated. The review identifies priorities for empirical evaluation, including multimodal benchmarks, standardized AL protocols, spatially and temporally independent validation, explainability, and continuous human-in-the-loop adaptation. This synthesis provides a methodological basis for future domain-adaptive LULC monitoring in the Brazilian Cerrado.

The stochastic nature of Large Language Models (LLMs) challenges traditional evaluation paradigms, which rely on single-response metrics and often mask complex behavioral patterns. This paper introduces Trait and Consistency Evaluation for LLMs (TraCE-LLM), an evaluation protocol that quantifies latent behavioral traits and model consistency within a black-box paradigm. Through a factorial design combining five LLMs, three benchmarks and a systematic stratification by prompt style (Naive, Chain-of-Thought and Adversarial), the framework employs a multidimensional rubric to measure Depth of Reasoning (DoR) and Originality (ORI) of model responses. The primary empirical contribution of this study is the identification and formalization of the Adversarial Compensation Effect (ACE), a phenomenon wherein smaller-capacity models under adversarial stress exhibit a paradoxical gain in accuracy metrics while suffering a severe degradation in behavioral stability. Our results also demonstrate an asymmetric stability with DoR being a significantly more stable trait than ORI and the prevalence of compressed reasoning, where 17.8% of correct answers lack adequate justification. By decoupling response correctness from process quality, TraCE-LLM provides a blueprint for more granular and reliable evaluation, arguing that LLM auditing must be multidimensional, context-sensitive and psychometrically informed to ensure the development of safer and more interpretable AI.

The development of urban air mobility (UAM) systems requires scalable, regulation-aware planning of low-altitude airspace and supporting infrastructure. Th