Abstract
Container terminals are of pivotal importance to global trade, as they act as a bridge between maritime and land transport. However, inefficiencies in operations, such as long waiting times and high emissions, continue to challenge the industry. Current practices, including first-come-first-served (FCFS) berth allocation, often result in ships arriving too early and idling at anchorage, leading to increased fuel consumption and negative environmental impacts. Just-in-Time (JIT) strategies have been identified as a potentially effective approach to address these issues by aligning ship arrivals with berth availability, thus optimising speed and reducing emissions. In this work, we present a simulation-based decision-support tool to evaluate JIT strategies in container terminal operations. By analysing scenarios involving speed optimisation and resource investments, the tool provides insights into key performance metrics, including waiting times, emissions, and resource utilisation. A case study designed around a large Portuguese seaport was used to validate the approach, with significant reductions in emissions and operational inefficiencies. These findings highlight the potential of JIT operations to enhance sustainability and efficiency in the maritime sector. Copyright © 2025. Published by Elsevier B.V.

Abstract
Circular startups (CSUs) play a crucial role in the circular transition by developing circular business models (CBMs) that minimise resource use and narrow material and energy loops. However, empirical research on how CBMs shape growth strategies and how ecosystems enable or constrain scaling remains limited. This study aims to fill this gap by analysing the growth strategy of CSUs, addressing their circularity, business model and scalability strategies. It analysed 44 CSUs operating in packaging and plastics, textiles and food, water and nutrients value chains, using a qualitative multiple-case design. Results show that CSUs predominantly adopt Commercial and ecosystem scalability strategies, linking replication and geographical expansion with access to partners, resources and markets, and implementing platform- or waste-based CBMs. The study expands existing frameworks by conceptualising Ecosystem Strategy as a core scalability approach and clarifying its mechanisms, offering guidance for entrepreneurs and policymakers seeking to foster circular transformation.

Abstract
Depression remains a major global mental health challenge, and early screening still depends largely on self-report questionnaires, despite their vulnerability to response bias and limited insight into the response process. This study examined the feasibility of a human-in-the-loop, LLM-based chatbot for depression screening under real-time supervision by a licensed psychologist. Twenty-one participants completed the DASS-21 in two digital formats: a static form and a conversational chatbot interface. In the chatbot condition, LLM-generated follow-up questions were reviewed by the psychologist before delivery. Interaction data were collected in both conditions using a digital phenotyping screening tool. The results suggest that the chatbot is a usable alternative for administering screening instruments and that process-level behavioral traces may add contextual information beyond final questionnaire responses. These findings support the potential of clinician-supervised chatbot workflows for enriching depression screening, while highlighting the need for further validation of the clinical relevance of such behavioral signals. © 2026 Copyright held by the owner/author(s).

Abstract

Abstract
Companies are increasingly adopting technologies such as Robotic Process Automation (RPA) to reduce costs and improve productivity. RPA is deployed in areas like accounting, payroll, and finance to automate business processes. While RPA does not necessarily result in unemployment, it has notable effects on employees and company governance. This study explores the impact of RPA implementation on employees and company governance, using a qualitative methodology based on thirteen semi-structured interviews with RPA experts from four multinational companies. The results indicate that the impacts of RPA vary depending on the automation strategy adopted (task-oriented or process-oriented). In task-oriented strategies, citizen developers often play a central role, contributing to rapid implementation. In contrast, process-oriented strategies tend to rely on professional developers and require more structured governance. The findings also point out that RPA influences not only task execution but also employee upskilling, job role redefinition, and the evolution of governance models. The study proposes an integrated framework linking automation strategy, governance, upskilling, and employee adaptation, offering both practical insights and theoretical contributions to digital transformation research and for managing risks and enhancing workforce capabilities. It also advances academic understanding by linking real-world RPA implementations to organisational and technological impacts.

Abstract
Large Language Models (LLMs) have become an essential tool for providing assistance, including in the form of virtual assistants that support everyday tasks. However, recent evidence suggests that LLMs can exhibit confirmation bias, responding in ways that reinforce beliefs and assumptions. This behavior negatively influences clinical decision-making processes within Clinical Decision Support Systems (CDSS), potentially compromising diagnostic accuracy and treatment recommendations. This study aims to quantify the prevalence of confirmation bias in LLMs, the conditions under which it occurs, and its causes. The methodology consisted of an experiment in which 52 biased, health-related queries were submitted to ten of the most widely used LLM models, and the responses were evaluated for bias. Statistical analyses were performed to indicate the presence of confirmation bias. Specifically, our results show that confirmation bias is present in all of the LLMs tested, with a prevalence ranging from 15% to 67%. Our findings suggest that LLMs' biases arise from factors such as the training dataset, the Transformer architecture's inherent characteristics, the prompt engineering and fine-tuning instructions added by developers. This study contributes to ongoing efforts towards the development of trustworthy LLMs. © 2026 IEEE.