In agile software development, user stories are a common tool for capturing functional requirements, valued for their simplicity and user-centric approach. Nevertheless, their inherent informality can introduce ambiguity, especially when acceptance criteria are lacking or poorly defined. Such ambiguity may hinder early design activities, including the development of UML robustness diagrams. This study investigates the impact of acceptance criteria and developer experience on the accuracy and efficiency of robustness diagram construction.A controlled experiment was conducted with thirty participants, divided into newbie and advanced developers, who were tasked with creating robustness diagrams from user stories both with and without acceptance criteria. Performance was assessed based on task duration and diagram completeness. Statistical analysis (Mann-Whitney U test) revealed that acceptance criteria significantly reduce errors and improve completion times, particularly among newbie developers. Experienced developers consistently produced higher-quality diagrams more efficiently, underscoring the role of expertise as a moderating factor.The findings suggest that well-specified acceptance criteria mitigate ambiguity, facilitating more accurate requirements interpretation and improving early design outcomes. Additionally, the results highlight the value of structured requirement practices in agile methodologies, especially in teams with varying levels of experience. This research advances the understanding of how requirement clarity and developer expertise collectively influence software modeling, providing actionable recommendations for enhancing agile design processes.

Este artículo tiene como objetivo diseñar una solución a un problema de pago y gestión de citas que tienen algunos Centros de Electrodiagnóstico y Fisioterapia. Esto se debe a que aún no se ha encontrado un sistema para su gestión administrativa. En consecuencia, realizar el proceso de alta de un paciente, agendar una cita o poner las notas de una consulta se hace de manera ineficiente utilizando hojas de cálculo. Diseñamos una aplicación web con la que los usuarios pueden realizar todo el proceso de gestión administrativa, como agendar citas, agregar un recordatorio de cita en Google Calendar, entre otras operaciones. Para la recolección de información utilizamos entrevistas semiestructuradas con los actores del proceso. En cuanto al análisis e interpretación de la información, se utilizaron técnicas, métodos y herramientas para diseñar diagramas e interfaces de usuario de manera formal. Con esto, el diseño de la solución propuesta puede ser beneficioso para este tipo de centros ya que la arquitectura se basó en las experiencias obtenidas de los stakeholders.

User stories are central to agile requirements engineering, but in practice they often exhibit ambiguity, missing information, and weak or absent acceptance criteria, thereby increasing rework and reducing verifiability. This paper presents LOONA, a mathematically formulated NLP/LLM decision-support pipeline for Spanish user story quality assurance. We model the task as supervised binary classification followed by conditional text generation: a fine-tuned Spanish BERT classifier estimates whether a story requires improvement, and a generative model is triggered only when the estimated class exceeds a decision threshold. The study uses an anonymized reconstruction of 100 Spanish-language user stories from an administrative software backlog, labeled by three agile requirements experts through majority vote. The reconstructed annotation table yields Fleiss’ $\kappa =0.76$, indicating substantial agreement. We report stratified 5-fold classification results, lexical baselines, rule-based baselines, automatic rewrite-similarity metrics, and paired expert review of held-out outputs. LOONA achieved Accuracy $=0.91$, Precision $=0.90$, Recall $=0.92$, F1-score $=0.91$, and MCC $=0.82$, outperforming TF-IDF logistic regression, TF-IDF support vector machines, and a rule-based baseline. In paired expert review ($n=30$), 70% of generated rewrites improved over the original story, 20% were unchanged, and 10% degraded; generated acceptance criteria were judged correct in 86.7% of cases and free of hallucinated information in 93.3%. The contribution is framed as a formal and empirical proof of concept for model-level LLM-assisted requirements quality assurance, with explicit limitations regarding dataset size, confidentiality, and human oversight.