We introduce FinMMDocR, a novel bilingual multimodal benchmark for evaluating multimodal large language models (MLLMs) on real-world financial numerical reasoning. Compared to existing benchmarks, our work delivers three major advancements. (1) Scenario Awareness: 57.9% of 1,200 expert-annotated problems incorporate 12 types of implicit financial scenarios (e.g., Portfolio Management), challenging models to perform expert-level reasoning based on assumptions; (2) Document Understanding: 837 Chinese/English documents spanning 9 types (e.g., Company Research) average 50.8 pages with rich visual elements, significantly surpassing existing benchmarks in both breadth and depth of financial documents; (3) Multi-Step Computation: Problems demand 11-step reasoning on average (5.3 extraction + 5.7 calculation steps), with 65.0% requiring cross-page evidence (2.4 pages average). The best-performing MLLM achieves only 58.0% accuracy, and different retrieval-augmented generation (RAG) methods show significant performance variations on this task. We expect FinMMDocR to drive improvements in MLLMs and reasoning-enhanced methods on complex multimodal reasoning tasks in real-world scenarios.

Significant tool support exists for the development of mixed reality (MR) applications; however, there is a lack of tools for analyzing MR experiences. We elicit requirements for future tools through interviews with 8 university research, instructional, and media teams using AR/VR in a variety of domains. While we find a common need for capturing how users perform tasks in MR, the primary differences were in terms of heuristics and metrics relevant to each project. Particularly in the early project stages, teams were uncertain about what data should, and even could, be collected with MR technologies. We designed the Mixed Reality Analytics Toolkit (MRAT) to instrument MR apps via visual editors without programming and enable rapid data collection and filtering for visualizations of MR user sessions. With MRAT, we contribute flexible interaction tracking and task definition concepts, an extensible set of heuristic techniques and metrics to measure task success, and visual inspection tools with in-situ visualizations in MR. Focusing on a multi-user, cross-device MR crisis simulation and triage training app as a case study, we then show the benefits of using MRAT, not only for user testing of MR apps, but also performance tuning throughout the design process.