arXiv:2601.11957v4 Announce Type: replace 
Abstract: Overlapping calendar invitations force busy professionals to repeatedly decide which meetings to attend, reschedule, or decline. We refer to this preference-driven decision process as calendar conflict resolution. Automating this decision process is crucial yet challenging. Scheduling logistics can drain hours, and human delegation often fails at scale, which motivates us to ask: Can we trust large language models (LLMs) or language agents to manage time? To enable a systematic study of this question, we introduce CalConflictBench, a benchmark for long-horizon calendar conflict resolution. In CalConflictBench, conflicts are presented to agents round-by-round over a calendar year, requiring them to infer and adapt to user preferences progressively. Our experiments show that current LLM agents perform poorly with high error rates, e.g., Qwen-3-30B-Think has an average error rate of 35%. To address this gap, we propose PEARL, a reinforcement-learning framework that (i) augments the language agent with an external preference memory that stores and updates inferred strategies (e.g., attendee priorities, topic importance, time/location preferences), and (ii) optimizes the agent with round-wise rewards that directly supervise decision correctness, ranking quality, and memory usage across rounds. Experiments on CalConflictBench show that PEARL achieves an error reduction rate of 0.76 and a 55% improvement in average error rate compared to the strongest baseline.

تم تقديم إطار عمل جديد يسمى PEARL لأتمتة حل نزاعات التقويم للمهنيين المشغولين، حيث يتناول التحديات التي تطرحها الدعوات المتداخلة. يستخدم هذا النظام التعلم المعزز ويتم تقييمه من خلال معيار يسمى CalConflictBench، الذي يقيم قدرة الوكلاء على التكيف مع تفضيلات المستخدمين بمرور الوقت. تظهر النماذج اللغوية الكبيرة الحالية معدلات خطأ مرتفعة في هذه المهمة، مما يبرز الحاجة إلى حلول محسنة.

Se ha introducido un nuevo marco llamado PEARL para automatizar la resolución de conflictos de calendario para profesionales ocupados, abordando los desafíos que plantean las invitaciones superpuestas. Este sistema utiliza el aprendizaje por refuerzo y se evalúa a través de un benchmark llamado CalConflictBench, que mide la capacidad de los agentes para adaptarse a las preferencias del usuario a lo largo del tiempo. Los modelos de lenguaje actuales muestran altas tasas de error en esta tarea, lo que resalta la necesidad de soluciones mejoradas.

Un nouveau cadre nommé PEARL a été introduit pour automatiser la résolution des conflits de calendrier pour les professionnels occupés, abordant les défis posés par les invitations qui se chevauchent. Ce système utilise l'apprentissage par renforcement et est évalué à travers un benchmark appelé CalConflictBench, qui évalue la capacité des agents à s'adapter aux préférences des utilisateurs au fil du temps. Les modèles de langage actuels montrent des taux d'erreur élevés dans cette tâche, soulignant la nécessité de solutions améliorées.

A new framework named PEARL has been introduced to automate calendar conflict resolution for busy professionals, addressing the challenges posed by overlapping invitations. This system utilizes reinforcement learning and is evaluated through a benchmark called CalConflictBench, which assesses agents' ability to adapt to user preferences over time. Current large language models (LLMs) show high error rates in this task, highlighting the need for improved solutions.