<p>Augmented reality (AR) has emerged as a human-centered technology for supporting manual assembly and maintenance by embedding digital information directly within the physical workspace. In manufacturing and technical training, AR may be proposed as an alternative to traditional two-dimensional documentation for conveying spatially complex instructions. Despite growing industrial interest, prior empirical studies report mixed evidence on whether head-mounted AR systems consistently outperform paper-based instructions, particularly as task complexity increases. This paper presents a controlled experimental comparison of head-mounted AR and paper-based diagrams during the assembly of a multi-component device. Fifty-six participants completed the same assembly procedure under one of the two instructional modalities. Performance was evaluated using decision efficiency, assembly accuracy, perceived workload, and system usability. Results show that AR enabled faster and more consistent decision making and reduced assembly errors, especially for steps involving complex spatial reasoning, while perceived workload remained comparable across conditions. Analysis of spatial interaction behavior further suggests that AR encourages use of a larger physical workspace. These findings indicate that spatially aligned AR can improve decision performance in assembly tasks where two-dimensional documentation is insufficient.</p>